data.jsonl

{"0": 1, "1": "techcrunch", "2": "https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/", "3": "Starcloud raises $250 million for orbital data centers as launch options dry up", "4": "Starcloud raises $250 million for orbital data centers as launch options dry up. There's about to be a big fight to secure access to space.", "5": "2026-08-22T18:05:06.652324"}
{"0": 2, "1": "techcrunch", "2": "https://techcrunch.com/podcast/the-doj-is-investigating-a16z-what-does-this-mean-for-venture-capital/", "3": "The DOJ is investigating a16z. What does this mean for venture capital?", "4": "The DOJ is investigating a16z. What does this mean for venture capital?. Andreessen Horowitz has two partners sitting on the boards of companies that now compete with each other: Ben Horowitz at Databricks and Martin Casado at Fivetran. Nothing too scandalous on the surface, except the Department of Justice has reportedly been investigating the arrangement for almost a year, dusting off a 112-year-old antitrust law that’s rarely used against VCs.  Board conflicts aren’t exactly new, and these companies weren’t necessarily direct competitors when a16z first invested […]", "5": "2026-08-22T18:05:06.656554"}
{"0": 3, "1": "techcrunch", "2": "https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/", "3": "AI data startup Micro1 reaches $500M gross run rate amid AI training boom", "4": "AI data startup Micro1 reaches $500M gross run rate amid AI training boom. Surging demand for AI training data is driving rapid growth for the startup and its rivals.", "5": "2026-08-22T18:05:06.662341"}
{"0": 4, "1": "techcrunch", "2": "https://techcrunch.com/2026/08/20/openai-is-gaining-on-anthropic-with-business-users-new-data-indicates/", "3": "OpenAI is gaining on Anthropic with business users, new data indicates", "4": "OpenAI is gaining on Anthropic with business users, new data indicates. Businesses are willing to flop back and forth as each lab releases new models, volatility that should give both companies' investors pause about how \"sticky\" enterprise AI spending really is.", "5": "2026-08-22T18:05:06.666369"}
{"0": 5, "1": "sifted", "2": "https://sifted.eu/articles/legal-dispute-monaco-bank/", "3": "Exclusive: Legal dispute emerges at Monzo cofounder's Monaco banking venture", "4": "Exclusive: Legal dispute emerges at Monzo cofounder's Monaco banking venture. ", "5": "2026-08-22T18:05:07.776707"}
{"0": 6, "1": "sifted", "2": "https://sifted.eu/articles/uk-chip-startup-fractile-in-talks-to-raise-at-6-5bn-valuation-reports-say/", "3": "UK chip startup Fractile in talks to raise at $6.5bn valuation, reports say", "4": "UK chip startup Fractile in talks to raise at $6.5bn valuation, reports say. ", "5": "2026-08-22T18:05:07.781629"}
{"0": 7, "1": "sifted", "2": "https://sifted.eu/articles/domyn-raises-over-1bn/", "3": "AI model maker Domyn raises over $1bn", "4": "AI model maker Domyn raises over $1bn. ", "5": "2026-08-22T18:05:07.787101"}
{"0": 8, "1": "sifted", "2": "https://sifted.eu/articles/callosum-raise-atomico-plural-uk-sovereign-ai/", "3": "Callosum raises $100m seed led by Atomico to tackle AI compute bottlenecks", "4": "Callosum raises $100m seed led by Atomico to tackle AI compute bottlenecks. ", "5": "2026-08-22T18:05:07.791936"}
{"0": 9, "1": "sifted", "2": "https://sifted.eu/articles/exclusive-ai-infrastructure-velatir-raises-e5m/", "3": "Exclusive: AI infrastructure startup Velatir raises \u20ac5m to accelerate AI adoption across Europe", "4": "Exclusive: AI infrastructure startup Velatir raises \u20ac5m to accelerate AI adoption across Europe. ", "5": "2026-08-22T18:05:07.796003"}
{"0": 10, "1": "sifted", "2": "https://sifted.eu/articles/domyn-uljan-sharka-podcast/", "3": "Domyn CEO Uljan Sharka: \u2018We're a few quarters away from $1bn ARR\u2019", "4": "Domyn CEO Uljan Sharka: \u2018We're a few quarters away from $1bn ARR\u2019. ", "5": "2026-08-22T18:05:07.802076"}
{"0": 11, "1": "sifted", "2": "https://sifted.eu/articles/revolut-ceo-to-borrow-250m/", "3": "Revolut to allow CEO to borrow up to $250m against his shares, reports say", "4": "Revolut to allow CEO to borrow up to $250m against his shares, reports say. ", "5": "2026-08-22T18:05:07.807032"}
{"0": 12, "1": "sifted", "2": "https://sifted.eu/articles/openai-clarifies-acquisition-offer/", "3": "OpenAI clarifies \u2018acquisition\u2019 offer for Irish teen's startup idea was a joke", "4": "OpenAI clarifies \u2018acquisition\u2019 offer for Irish teen's startup idea was a joke. ", "5": "2026-08-22T18:05:07.812413"}
{"0": 13, "1": "sifted", "2": "https://sifted.eu/articles/revolut-nik-storonsky-quantumlight-closes-500/", "3": "Exclusive: Nik Storonsky\u2019s VC firm closes $500m for second fund", "4": "Exclusive: Nik Storonsky\u2019s VC firm closes $500m for second fund. ", "5": "2026-08-22T18:05:07.819047"}
{"0": 14, "1": "theinfo", "2": "https://www.theinformation.com/articles/america-hates-data-centers", "3": "America Really, Really Hates Data Centers", "4": "America Really, Really Hates Data Centers. <p>\u2022 <i>The Big Read: </i>AI promises to cure cancer. Scientists <a href=\"https://www.theinformation.com/articles/ai-probably-cure-cancer-anytime-soon-scientists-say\">feel existential dread</a></p><p>\u2022 The new robotics \u2018<a href=\"https://www.theinformation.com/articles/new-robotics-arms-race-can-craziest-hype-video?rc=sslhyj\">arms race</a>\u2019: Who can do the craziest hype video?</p><p>\u2022 <i>Plus, Recommendations\u2014our weekly pop culture picks: </i>\u201c<a href=\"https://www.audible.com/podcast/Dan-Taberskis-Manifesto/B0H5TR6GRP?srsltid=AfmBOoogn-4JU1BZYJX7ivfYTeIuykSkEYaaL-q_jT0yvXXydLX_Vh-s\">Dan Taberski\u2019s Manifesto</a>,\u201d \u201c<a href=\"https://www.amazon.com/Tender-Age-Novel-Chang-rae-Lee/dp/B0G2GGTQDL\">A Tender Age</a>\u201d and &nbsp;\u201c<a href=\"https://www.rottentomatoes.com/m/the_end_of_oak_street\">The End of Oak Street</a>\u201d </p><hr /><p><b>With surprising swiftness</b>, data centers have become the b\u00eate noire of the left, the right and seemingly everyone between those two poles. That poses an immense problem for Silicon Valley, even if it is an issue partially of the industry\u2019s own making.&nbsp;</p><p>My colleagues at The Information have been following the mounting opposition to data centers with <a href=\"https://www.theinformation.com/articles/data-center-bans-top-500-new-york-texas-join-pushback?rc=sslhyj\">our data center moratorium tracker</a> for several months. And an amazing array of additional examples has recently arrived that further illustrate how widespread the sentiment has grown. Weeks ago, Greg Abbott, the Republican governor of anything-goes Texas, <a href=\"https://www.wsj.com/politics/policy/politicians-who-once-championed-data-centers-are-now-bashing-them-c172d4cb\">brought the construction</a> of 1,800 data centers in that state to a standstill over concerns about their power and water use. And then on Tuesday, Josh Shapiro, the Democratic governor of Pennsylvania, <a href=\"https://www.inquirer.com/politics/pennsylvania/josh-shapiro-data-center-order-20260818.html\">issued an executive order</a> that tightly restricts data center expansion in his state.&nbsp;</p><p>Then there was <a href=\"https://www.youtube.com/watch?v=r9OhekhlZ9Y\">the pee ad</a> that went viral on Wednesday from Liquid Death and the beer company owned by ex-NFL stars Travis and Jason Kelce: The two startups joined up to cajole people to consume copious amounts of weak pilsner and canned water, urinate and then mail off their bodily fluids to a stockpile, which they promise to share with data centers to use in their liquid-cooling systems. They\u2019re kidding, I think, but I can\u2019t entirely be sure: Neither Liquid Death nor the Kelce brothers\u2019 Garage Beer would comment for this story. (According to TechCrunch, <a href=\"https://techcrunch.com/2026/08/20/ok-can-we-actually-cool-data-centers-with-our-pee/\">it\u2019s theoretically possible</a> to use urine to cool data centers. Take that as you will.)&nbsp;</p><p>Fresh <a href=\"https://heatmap.news/daily/data-center-opposition-poll-collapse\"", "5": "2026-08-22T18:05:08.105346"}
{"0": 15, "1": "theinfo", "2": "https://www.theinformation.com/briefings/tiktok-reaches-400-million-settlement-doj-child-privacy-lawsuit", "3": "TikTok Reaches $400 Million Settlement With DOJ Over Child Privacy Lawsuit", "4": "TikTok Reaches $400 Million Settlement With DOJ Over Child Privacy Lawsuit. <p>The Department of Justice <a href=\"https://www.justice.gov/opa/pr/justice-department-secures-400m-settlement-tiktok-and-bytedance-resolve-childrens-privacy\">said</a> Friday that TikTok and its parent company ByteDance have agreed to pay $400 million to settle a lawsuit regarding their alleged violation of federal children\u2019s privacy laws.</p> <p>The DOJ originally filed its complaint in 2024, alleging that TikTok had violated the ...</p>", "5": "2026-08-22T18:05:08.119305"}
{"0": 16, "1": "theinfo", "2": "https://www.theinformation.com/briefings/exclusive-openai-explored-stake-stargate-power-developer-lancium", "3": "Exclusive: OpenAI Explored Stake in Stargate Power Developer Lancium", "4": "Exclusive: OpenAI Explored Stake in Stargate Power Developer Lancium. <p>OpenAI explored its own investment in or acquisition of Lancium, the power developer behind its Stargate facility in Abilene, Tex., last quarter, <a href=\"https://www.theinformation.com/newsletters/ai-infrastructure/nvidia-using-land-electricity-deals-lock-hardware-bundle?rc=zjctrx\">The Information reported Friday</a>. Ultimately, Nvidia <a href=\"https://www.theinformation.com/articles/nvidia-invest-3-billion-blackstone-backed-power-firm-behind-stargate?rc=zjctrx\">invested billions of dollars in the firm</a>.</p> <p>The OpenAI interest shows the length to which the AI ...</p>", "5": "2026-08-22T18:05:08.130785"}
{"0": 17, "1": "theinfo", "2": "https://www.theinformation.com/articles/nvidia-using-land-electricity-deals-lock-hardware-bundle", "3": "Nvidia is Using Land and Electricity Deals to Lock In Its Hardware Bundle", "4": "Nvidia is Using Land and Electricity Deals to Lock In Its Hardware Bundle. <p>Nvidia is racing to lock up data center capacity for its AI hardware before its rivals do. On Friday, the company announced it had acquired a minority stake in Cloverleaf Infrastructure, its third equity investment in rapid succession in firms that secure real estate and electricity access for data centers in the U.S.</p><p>By backing the two-year-old Cloverleaf Infrastructure alongside recent stakes in <a href=\"https://www.theinformation.com/articles/nvidia-invest-3-billion-blackstone-backed-power-firm-behind-stargate?rc=c48ukx\">Lancium</a> and <a href=\"https://www.theinformation.com/articles/nvidia-talks-invest-3-billion-sb-energy-part-openai-data-center-deal?utm_campaign=article_email&amp;utm_content=article-17625&amp;utm_medium=email&amp;utm_source=sg&amp;rc=c48ukx\">SB Energy</a>, Nvidia is locking in near-term power rights for facilities that could use multiple generations of its chips over many years. That positioning can also help Nvidia head off <a href=\"https://www.theinformation.com/newsletters/ai-infrastructure/fearing-ai-chip-glut-data-center-developers-choosin-texas?rc=c48ukx\">what some data center developers fear is a looming chip overhang</a> in which energy grid delays and local political pushback leave new AI hardware sitting idle without electricity.</p>", "5": "2026-08-22T18:05:08.136720"}
{"0": 18, "1": "theinfo", "2": "https://www.theinformation.com/briefings/dragoneer-founder-stad-buy-timberwolves-controlling-stake", "3": "Dragoneer Founder Stad to Buy Timberwolves Controlling Stake", "4": "Dragoneer Founder Stad to Buy Timberwolves Controlling Stake. <p>Marc Stad, the founder of Dragoneer Investment Group, is buying a controlling stake in the Minnesota Timberwolves and Minnesota Lynx professional basketball teams, according to The New York Times\u2019 Athletic publication.</p> <p>Stad is buying the interest at a $4.5 billion valuation from Marc Lore, the ...</p>", "5": "2026-08-22T18:05:08.143841"}
{"0": 19, "1": "theinfo", "2": "https://www.theinformation.com/articles/barclays-loses-tech-bankers-vcs-get-exits-poolside-openrouter", "3": "Barclays Loses Tech Bankers; VCs Get Exits From Poolside, OpenRouter", "4": "Barclays Loses Tech Bankers; VCs Get Exits From Poolside, OpenRouter. <p>Barclays has been losing tech bankers lately.&nbsp;</p><p>Last night I <a href=\"https://www.theinformation.com/briefings/veteran-barclays-semiconductor-banker-tim-luke-join-morgan-stanley?rc=zjctrx\">reported that Tim Luke</a>, a veteran semiconductor analyst and banker, is leaving the British investment bank to join Morgan Stanley. His position as a vice chairman in technology investment banking will further strengthen Morgan Stanley\u2019s strong semiconductor banking franchise.&nbsp;</p><p>Not too long ago, semiconductor dealmaking was overlooked by Wall Street, as software and internet matchmaking dominated headlines. But that\u2019s changed thanks to the importance of chips in training and running AI. An increase in semiconductor M&amp;A plus the initial public offerings of companies like Cerebras have put a premium on rainmakers with long-standing networks and knowledge in the industry.&nbsp;&nbsp;</p>", "5": "2026-08-22T18:05:08.149427"}
{"0": 20, "1": "theinfo", "2": "https://www.theinformation.com/briefings/nvidia-reportedly-pay-6-billion-licensing-hiring-deal-ai-model-startup-poolside", "3": "Nvidia to Reportedly Pay $6 Billion in Licensing and Hiring Deal with AI Model Startup Poolside", "4": "Nvidia to Reportedly Pay $6 Billion in Licensing and Hiring Deal with AI Model Startup Poolside. <p>Nvidia has agreed to pay $6 billion to license AI model-development software from startup Poolside, the startup told investors in a letter first reported by <a href=\"https://www.newcomer.co/p/sources-poolside-strikes-6-billion?utm_campaign=email-post&amp;r=a9xr6&amp;utm_source=substack&amp;utm_medium=email\">Newcomer</a>.</p> <p>Poolside was an early developer of a coding AI agent and pivoted to developing data centers&nbsp;before releasing its own open ...</p>", "5": "2026-08-22T18:05:08.157092"}
{"0": 21, "1": "theinfo", "2": "https://www.theinformation.com/briefings/walmarts-u-s-sales-slowdown-sends-stock-tumbling-9", "3": "Walmart\u2019s U.S. Sales Slowdown Sends Stock Tumbling 9%", "4": "Walmart\u2019s U.S. Sales Slowdown Sends Stock Tumbling 9%. <p>Shares of Walmart fell 9% after the retail giant reported a sharp slowdown in growth of comparable sales at Walmart U.S., even as Walmart\u2019s online sales continued to grow strongly. Overall, Walmart\u2019s revenue grew 5.9% to $187.9 billion.</p> <p>Walmart said comparable sales at its U.S. business grew ...</p>", "5": "2026-08-22T18:05:08.163021"}
{"0": 22, "1": "theinfo", "2": "https://www.theinformation.com/articles/robots-gpt-2-era", "3": "Robots Are in Their GPT-2 Era", "4": "Robots Are in Their GPT-2 Era. <p>It\u2019s no secret that AI-powered robots aren\u2019t so good yet. They struggle with a broad range of simple tasks, from untangling cables to chopping vegetables. Nonetheless, morale is high among roboticists who are flush with venture cash as they work toward a \u201c<b>ChatGPT moment</b>\u201d when robots become broadly useful and available.&nbsp;</p><p>Until then roboticists are happy to celebrate little milestones. I spent two days this week at <b>Actuate</b>, a robotics conference in San Francisco, along with 1,200 founders, developers and engineers, where I saw demonstrations and video footage showing signs of progress.</p><p>When <b>Chelsea Finn</b>, a Stanford University professor of computer science and electrical engineering, played a clip of a mechanical robot arm successfully making a latte, the audience broke into applause\u2014no matter that it moved slowly and a human had to steam the milk. (Some robot baristas are already serving coffee, for example at the SFO airport, but software code controls their actions, rather than a single AI model like PI\u2019s that is trained to perform a variety of tasks.)</p><p>Finn, who is also a co-founder of robotics software firm <b>Physical Intelligence</b>, said her field resembles language AI in the era following OpenAI\u2019s release of <b>GPT-2</b> in 2019, when large language models were beginning to perform a variety of tasks without being specially trained or fine-tuned to specific ones. <a href=\"https://www.pi.website/blog/pi07\">PI\u2019s latest model</a> similarly works out of the box on some limited tasks, such as using a pair of robot arms with pinchers to fold laundry and cardboard boxes.</p>", "5": "2026-08-22T18:05:08.171318"}
{"0": 23, "1": "theinfo", "2": "https://www.theinformation.com/briefings/alibaba-ceo-expects-ai-related-arr-reach-10-billion-september", "3": "Alibaba CEO Expects AI-Related ARR to Reach $10 Billion by September", "4": "Alibaba CEO Expects AI-Related ARR to Reach $10 Billion by September. <p>Alibaba Group CEO Eddie Wu said on an earnings call Thursday that the company expects the annualized revenue run rate for its AI-related products to reach $10 billion in the current quarter through September, up from $7.3 billion in the previous quarter.</p> <p>Alibaba\u2019s revenue in the quarter through ...</p>", "5": "2026-08-22T18:05:08.180178"}
{"0": 24, "1": "theinfo", "2": "https://www.theinformation.com/briefings/exclusive-ubs-hires-jp-morgan-banker-su-ramp-ai-banking-efforts", "3": "Exclusive: UBS Hires JP Morgan Banker Su to Ramp Up AI Banking Efforts", "4": "Exclusive: UBS Hires JP Morgan Banker Su to Ramp Up AI Banking Efforts. <p>UBS is hiring Nicole Su from JPMorgan to lead its AI investment banking efforts, according to an internal memo reviewed by The Information.</p> <p>Su, most recently an executive director at JPMorgan, will join the bank later this month as its head of emerging AI technology banking in the Americas, ...</p>", "5": "2026-08-22T18:05:08.184603"}
{"0": 25, "1": "theinfo", "2": "https://www.theinformation.com/articles/anthropics-enterprise-ai-venture-buys-consultancy", "3": "Anthropic\u2019s Enterprise AI Venture Buys Consultancy", "4": "Anthropic\u2019s Enterprise AI Venture Buys Consultancy. <p>A joint venture established by Anthropic and Wall Street firms such as Blackstone has made its first acquisition since its July launch as it looks to boost the growth of Claude, Anthropic\u2019s chatbot, among businesses.</p>\n\n<p><a href=\"https://www.theinformation.com/org-charts/anthropic\">Anthropic</a>\u2019s venture, called Ode with Anthropic, plans to announce Thursday that it is buying an AI consultancy, Casper Studios, which helps businesses develop applications using AI, according to Ode. It\u2019s making the acquisition at a time when corporate executives are trying to figure out how to get the most from AI and are wary of rising bills from the new technology.</p>", "5": "2026-08-22T18:05:08.190796"}
{"0": 26, "1": "theinfo", "2": "https://www.theinformation.com/briefings/veteran-barclays-semiconductor-banker-tim-luke-join-morgan-stanley", "3": "Veteran Barclays Semiconductor Banker Tim Luke to Join Morgan Stanley", "4": "Veteran Barclays Semiconductor Banker Tim Luke to Join Morgan Stanley. <p>Tim Luke, a vice chairman in Barclays global technology investment banking group, is joining Morgan Stanley as a senior investment banker,&nbsp;according to people familiar with his move.</p> <p>Luke, a long-time semiconductor investment banker and equity research analyst, advised on some of the ...</p>", "5": "2026-08-22T18:05:08.196659"}
{"0": 27, "1": "hackernews", "2": "https://github.com/pawaca/dsh-edge", "3": "Show HN: Running a full AI coding agent inside Cloudflare Durable Object", "4": "Show HN: Running a full AI coding agent inside Cloudflare Durable Object. ", "5": "2026-08-22T18:05:09.037286"}
{"0": 28, "1": "hackernews", "2": "https://ozbrain.com", "3": "Show HN: OzBrain, a shared brain for knowledge between agents and your team", "4": "Show HN: OzBrain, a shared brain for knowledge between agents and your team. I think agent-first chat interfaces will be a primary software modality and busy dashboard&#x2F;UI will go away. I\u2019m not sure who exactly wins it, but I want my knowledge to grow&#x2F;go with me.<p>A lot of the \u201cknowledge\u201d ie research, analysis, reasoning will be done by agents as the primary user. Our current notes tools &amp; tasks management systems were built for humans\u2026 I don\u2019t care what the 17th thing on my bug backlog is. I want to conduct agents that can execute for me and do great work.<p>What I built OzBrain to do:\n+ Create a central place for agent reasoned knowledge to live\n+ Be agnostic about what apps&#x2F;agents connect to it\n+ Capture everything and track it so I can audit it\n+ Enable teams, collaborators or partners to share brains\n+ Handle conflicts so many agents in the same article doesn\u2019t blow up\n+ Refactor knowledge into more token friendly chunks and map the index well\n+ Close the knowledge loop so new thinking supersedes old thinking across the corpus. Don\u2019t erase, depreciate and link\n+ Keep user data safe and secure\n++ Be easy enough to use that you don\u2019t have to have any technical knowledge<p>Some among us will always build their own custom solutions, but there are millions of tech professionals and small business owners that will use agents heavily and need a solution. So I\u2019m trying to build that.<p>Isn\u2019t this like gBrain? Yes, similar. I think it\u2019s like AWS vs Vercel. AWS is very powerful, configurable, and useful if you\u2019re technical and want to invest the time into really fine tuning your system\u2026 but if you just want your web deploy&#x2F;hosting to just work and be easy to deal with you use Vercel.<p>&#x2F;&#x2F; WHY I MADE IT<p>I\u2019ve been enjoying getting back to my technical roots, as I lost my coding skills more than a decade ago, but with AI I can focus on the system and the product in partnership with agent coding workflows.<p>I recently built a Voice AI for older people. To build it I created an agentic engineering workflow (feel free to rip that up as I\u2019m always looking to improve systems: <a href=\"https:&#x2F;&#x2F;ozbrain.com&#x2F;resources&#x2F;eng-flow\" rel=\"nofollow\">https:&#x2F;&#x2F;ozbrain.com&#x2F;resources&#x2F;eng-flow</a>) My approach with coding agents is trust but verify, and I\u2019m trying to replace the parts where a human would review with an adversarial or specialized agent who would give a better answer&#x2F;review.<p>I have workflows that will go high level task to shipped PR running in Claude cloud sessions. I use Claude Code locally and Cursor when I want a tighter loop on doing visual work like UI or layout. And Codex to either load balance usage for TokenThriffting or when I want a different llm to think thru something.<p>It was a pain in the ass passing .md files around and keep track of which version was the most recent, so I built a hosted .md storage right in Supabase and any of my agents already have Supabase access. This let me build a solid, scalable, secure voice AI from my phone at the gym. All my agents have access to our knowledge, can write to it, update and refer to it as we build and improve the product and the systems we use.<p>Out of 75 founder friends I asked about how they manage shared knowledge, 26 built their own custom knowledge systems\u2026 Obsidian vaults with 7k files synced through a VPS, markdown repos behind their own MCP servers, cron jobs stitching Supabase to a skills file\u2026 each a different Frankenstein they have to maintain. 32 said they felt the pain of moving static files around but didn\u2019t have any solution for it.<p>So I rebuilt my brain better and used it to build it.<p>&#x2F;&#x2F; HOW YOU CAN HELP<p>Would love to have you try it out. The maintenance loop is still in alpha so not running it on customer data yet.<p>If you built your own brain I\u2019d love to hear how you did it. What criteria was most important for you in its design &amp; function.<p>If you are tired of shuffling .md files around I\u2019d love to have you try out OzBrain and to give feedback, just ask your agent to put it in the shared bugs &amp; features brain!<p>Cheers!\nBubs.co", "5": "2026-08-22T18:05:09.041997"}
{"0": 29, "1": "hackernews", "2": "https://traccia.ai/", "3": "Show HN: Traccia - Observability, Runtime Control & Audit for agents", "4": "Show HN: Traccia - Observability, Runtime Control & Audit for agents. AI applications are becoming agents, which has started to take autonomous decisions. There are plenty of tools and platform available to trace, and observe what an agent or llms calls does. They are good in what they do, but tracing and observability isnt enough for AI agents era. We need a solution that can help you observe, evaluate, create run time policies to govern and finally audit the actions of the agent. We built Traccia to solve this problem. The good part, all of these can be achieved by just writing few lines of code. Traccia has an open-sourced sdk that can work with your existing observability tool like grafana, tempo, jaeger, etc. In case you need more than just observability, Traccia provides the platform to evaluate, control and audit the agents. The platform is easy to use. The product&#x27;s documentation is quite extensive. It is also cloud vendor and framework agnostic. Traccia is being built by an Indian start up ,based out of Bengaluru. We are running a 3 months free trials so that you can explore without any strings attached. We are open to improvise and get better so please drop your comments and feedbacks.", "5": "2026-08-22T18:05:09.049428"}
{"0": 30, "1": "hackernews", "2": "https://aitoolsinsiderhq.com/log/", "3": "Show HN: An autonomous AI agent running one project for two months in public", "4": "Show HN: An autonomous AI agent running one project for two months in public. ", "5": "2026-08-22T18:05:09.055395"}
{"0": 31, "1": "hackernews", "2": "https://www.basecompute.co/local", "3": "Show HN: Zero () friction local AI for Mac", "4": "Show HN: Zero () friction local AI for Mac. Super excited to launch our new app Local today. What we\u2019ve learned at Base Compute over the last months is that running AI directly on your laptop or workstation gives you maximum privacy and it\u2019s free, but it\u2019s also a massive headache to configure. So we\u2019ve decided what matters is making the experience completely frictionless for users.<p>Local analyses the hardware of your laptop, optimises the AI for it, and recommends the best models for your specific device.<p>It let\u2019s you do what you\u2019re doing with cloud AI already, just for free and on your own machine: Chatting with PDF\u2019s, Recording and summarising meetings, running coding agents...<p>If you\u2019re using Local in your office with colleagues, you can run it in \u201cOffice Mode\u201d. The strongest computer in your office runs the AI and everyone can connect to it with their laptop. The data never leaves the office.<p>It\u2019s available for download on our website today, please try it out and let us know what you think!", "5": "2026-08-22T18:05:09.061298"}
{"0": 32, "1": "hackernews", "2": "https://www.danielvaughn.dev/posts/huzzah/", "3": "Show HN: Huzzah \u2013 a novel approach to coding with AI", "4": "Show HN: Huzzah \u2013 a novel approach to coding with AI. Hello everyone. I&#x27;ve been working on this experimental editor called Huzzah.<p>I&#x27;ve been working almost exclusively with coding agents since January of this year, and over the past few months I began to feel utterly exhausted by them. They&#x27;re great, but I&#x27;m finding it more and more tedious to write full sentences for every change I want. Not only that, but it seems there&#x27;s a complexity limit for codebases - beyond a certain point the agent begins confusing itself.<p>I&#x27;d like to go back to writing code, but I don&#x27;t want to go all the way back to fully manual coding. So I&#x27;ve come up with this interaction paradigm where you:<p><pre><code>  1. write pseudocode in whatever way makes the most sense to you\n  2. on save, the editor synchronizes your work to real source code\n  3. the pseudocode is persisted alongside the generated code, making your prompt effectively a stored record of intent.\n</code></pre>\nIt may not work for every use case, but in my initial playthroughs I&#x27;ve found it very enjoyable.<p>Right now it&#x27;s just a proof of concept - installation instructions are here in the readme: <a href=\"https:&#x2F;&#x2F;github.com&#x2F;danielvaughn&#x2F;hz\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;danielvaughn&#x2F;hz</a><p>You can also watch a video of it in action here: <a href=\"https:&#x2F;&#x2F;x.com&#x2F;danielvaughn&#x2F;status&#x2F;2090456808431165715\" rel=\"nofollow\">https:&#x2F;&#x2F;x.com&#x2F;danielvaughn&#x2F;status&#x2F;2090456808431165715</a><p>Cheers!", "5": "2026-08-22T18:05:09.067247"}
{"0": 33, "1": "hackernews", "2": "https://tablecanon.app/", "3": "Show HN: Building Table Canon, an AI Campaign Memory Engine for TTRPGs", "4": "Show HN: Building Table Canon, an AI Campaign Memory Engine for TTRPGs. Hey HN! I built Table Canon to solve a problem my playgroup kept running into: 3-4 hour tabletop gaming sessions leave behind massive audio recordings, but standard meeting note-takers treat every session as an isolated island, butcher fantasy terms, and don&#x27;t know who is speaking.<p>I wanted an engine that tracks long-term state across months of games, so I built a pipeline to extract entity updates, open quest hooks, and character promises across sessions.<p>The Tech Stack:<p>* Transcription: whisper-large-v3-turbo  \n* Diarization: pyannote for speaker embeddings &amp; voice profile matching  \nExtraction &amp; Memory: OpenAI API with Structured Outputs (JSON Schema enforcement for state updates)  \n* TTS &amp; Audio Recaps: Kokoro &#x2F; Chatterbox Turbo  \nMusic Generation: ACE-Step-v1.5-XL-Turbo for rendering session summaries into lyrics&#x2F;ballads<p>A Few Engineering Lessons &amp; Challenges:<p>* State Delta Extraction vs. Context Explosions: Feeding 20 prior session transcripts into context windows quickly becomes cost-prohibitive and noisy. Instead of re-reading raw history, each session outputs an atomic state delta (updates to NPC dossiers, new locations, resolved promises) to a database. Keeping context bounded as campaigns stretch past session 30+ has been one of the trickiest architectural hurdles.\n* Custom Pre-Lexicons: General STT models struggle with homebrew proper nouns (turning fantasy names into standard dictionary words). Injecting a pre-pass fantasy term dictionary into prompt context significantly improved first-pass spelling.\n* VAD &amp; Audio Chunking: Passing a 4-hour raw audio file directly to Pyannote&#x2F;Whisper leads to memory leaks and process timeouts. Pre-processing with Voice Activity Detection (VAD) and deterministic chunking was necessary before touching the models.<p>Current Limitations &amp; Active Hard Problems:<p>* Entity Alias Resolution: Matching entities across sessions when players use varying aliases or informal shorthand (e.g., matching &quot;The Red Bishop&quot; to &quot;Arthur&quot; or &quot;that cult leader guy&quot;) without accidentally merging distinct NPCs. I address this, partially, but allowing the user to Edit aliases, merge or split entities after-the-fact.\n* Quest &amp; Hook Resolution Logic: Fine-tuning the LLM to reliably determine whether a promise, open mystery, or quest has actually been resolved versus remaining open or implicitly abandoned.<p>I&#x27;d love feedback on how others are handling these sorts of issues - or any notes for folks who try it out! No initial login required with 6 hours of upload available to try.", "5": "2026-08-22T18:05:09.074419"}
{"0": 34, "1": "hackernews", "2": "https://epho.io", "3": "Show HN: Epho \u2013 run Claude Code with a curl", "4": "Show HN: Epho \u2013 run Claude Code with a curl. Hey folks, Burak here.<p>Epho is an API that allows running Claude Code, Codex or Opencode in a sandbox in the cloud. It abstracts away sandboxes, and allows running coding agents with a single HTTP request.<p>Epho came out of our own struggles with building our own AI analyst:\n- Sandboxes give you bare machines; you need to configure them for agentic workloads.\n- Each agent behaves differently, and you need to build integrations with each of them.\n- Sandbox providers are not very reliable, which means you need to figure out a multi-provider strategy to avoid failures.\n- Logging, artifacts, input&#x2F;output, event streaming, and all of the other operational aspects need to be figured out.<p>We had to go through the pain ourselves. We got to a point where things got quite reliable, and it became more obvious to us that this should be a primitive on its own: send a POST request, get the events streaming back to you.<p>Epho is an agents-as-an-API product: you send a request, it spins up a sandbox, configures the chosen harness, clones your repos, and kicks off the agent. It takes care of automatic fallbacks across different providers, handles auth stuff, and just streams back the events and outputs.<p>It supports Claude Code, Codex and Opencode out of the box, and pretty much all the models they support out of the box. It streams the events back, handles attachments and output files, automatically manages the fallbacks on different sandbox providers, retries, and all the auth stuff. You just send a prompt, your repo, MCP servers you want to use with it, and it runs them.<p>I recorded a demo here to show a real example: <a href=\"https:&#x2F;&#x2F;youtu.be&#x2F;HGfly1aytPA\" rel=\"nofollow\">https:&#x2F;&#x2F;youtu.be&#x2F;HGfly1aytPA</a><p>I am quite excited for Epho, simply because I think it is a new primitive that would allow building agents into product a lot easier than it is today. We are running our agents on Epho on prod, so we&#x27;ll keep maintaining it regardless, and we wanted to ship it as an independent product.<p>Epho is free to get started, and you can run it with Opencode&#x27;s free models to get started with it.<p>I am quite curious to hear what you&#x27;d think and would love to get your feedback.<p>Cheers,\nBurak", "5": "2026-08-22T18:05:09.084124"}
{"0": 35, "1": "hackernews", "2": "https://github.com/runvendo/vendo", "3": "Launch HN: Vendo (YC S26) \u2013 Let users build features on top of your product", "4": "Launch HN: Vendo (YC S26) \u2013 Let users build features on top of your product. Hi HN, we\u2019re Yousef &amp; Nour, founders of Vendo (<a href=\"https:&#x2F;&#x2F;vendo.run\">https:&#x2F;&#x2F;vendo.run</a>). Vendo lets users create new features inside the software they already use. A user describes the dashboard, workflow, or small app they need, and Vendo builds it on top of the product\u2019s existing data, API, and interface.<p>Demo: <a href=\"https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=VdpHehY64ls\" rel=\"nofollow\">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=VdpHehY64ls</a><p>We built Vendo because every SaaS eventually faces the same problem: every customer needs something slightly different. One wants a new report and another needs a workflow that only makes sense for their team. These requests either sit on the roadmap, become one-off engineering work, or force the customer into spreadsheets and external tools. We wanted the user to be able to create the missing feature themselves, without leaving the product.<p>Here is how it works:<p>- npx vendo init reads the product&#x27;s API surface, theme, routes, and more. These are used so that the apps Vendo creates (1) look on-brand and native and (2) have the ability to read data and perform actions directly through the company&#x27;s API<p>- When a user asks for a feature, we have a custom Vendo harness that writes a React component with a bunch of Vendo add-ons and guardrails (ex. ability to make calls to the host API + our component library). Every save is compiled, type-checked, run against real API responses, and rendered before the user sees it. We just released a benchmark and write-up here with more info for anyone interested: <a href=\"https:&#x2F;&#x2F;vendo.run&#x2F;blog&#x2F;generating-product-ui-measured\">https:&#x2F;&#x2F;vendo.run&#x2F;blog&#x2F;generating-product-ui-measured</a><p>- We use QuickJS to make sure that anything the agent creates is sandboxed and can&#x27;t mess with the company&#x27;s site. Vendo compiles the component and runs it with Preact inside a QuickJS VM with no access to the DOM, network, or clock. The VM returns a UI tree, which the host renders using the product\u2019s registered components. When the user clicks something, QuickJS emits a tool call; the host executes it through Vendo\u2019s guard and passes the result back into the same VM, preserving the screen\u2019s local state.<p>There&#x27;s a lot of generative UI right now: streaming developer-written components into a chat (Vercel AI SDK, CopilotKit, Thesys), or rendering your app inside someone else&#x27;s assistant (OpenAI Apps SDK, MCP Apps). We differ on two things. Vendo lives in your product and acts through your API as the signed-in user, so what it makes is durable: real apps users keep, pin, and run on triggers while they&#x27;re away, and not components that are merely confined to a chat. Plus, it&#x27;s not capped at putting together a bunch of prebuilt components: the agent can build arbitrary apps, from a quick dashboard out of your own components to real custom code running in a sandbox, and either way data only ever comes from tool calls to your API.<p>Here are some things customers are using Vendo for today:<p>- Letting their users create custom dashboards and reports. These are mainly UI-based and focused on letting the user see the exact graphs and metrics they care about<p>- Letting their customers create recurring automations. A big thing as well that has been used for these automations is the fact that we connect to external connections, so users have been automating many of their inter-tool workflows (ex. an automation that sends a slack alert based off of something in the product)<p>- B2B customers letting their customers customize the product with specific business logic. Often this is simple things like an extra field on a form, or an extra permission, but it is hard for a business to keep up with them otherwise.<p>- Creating and sharing custom dashboards&#x2F;apps across an organization. Since the apps Vendo creates are durable, they can be shared, reused, and forked (which can\u2019t be done with many of the other in-chat generative UI solutions)<p>We&#x27;ve spent a lot of time thinking about how AI and agents will change the way people consume software. We think the answer is personal(ized) software: you see the UI you need to see, you tell an agent exactly what you need, and the product molds to how you work.<p>The key insights that have enabled the product to work are:<p>- A rule in code always beats a rule in a prompt.<p>- Invent as little syntax as possible. Generation got faster and more reliable when the output looked like what models already know (JSX-shaped markup) instead of a clever custom format.<p>- Deterministic beats model wherever you can get away with it. Theme extraction is pure static analysis, and a remix starts as a copy of your component, no model call.<p>Vendo is completely open-source (Apache-2.0) and can be self-hosted, so feel free to c", "5": "2026-08-22T18:05:09.091842"}
{"0": 36, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49373445", "3": "Ask HN: What AI automations have you kept running in production?", "4": "Ask HN: What AI automations have you kept running in production?. ", "5": "2026-08-22T18:05:09.100214"}
{"0": 37, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49371913", "3": "Offline RAG on iOS with Spatial Integration", "4": "Offline RAG on iOS with Spatial Integration. I&#x27;m the developer behind CartoType. I\u2019ve been working on bridging local language models with offline mapping, and I have just put together a demo of a completely offline Spatial RAG pipeline running natively on an iPhone. The new system is named the CartoType Field Assistant. You can find the website at https:&#x2F;&#x2F;cartotype.com<p>Demo (1m 19s): https:&#x2F;&#x2F;www.youtube.com&#x2F;shorts&#x2F;a8yQPn7_jyI<p>Use cases: Any organisation with field technicians or emergency first responders needs complex procedural knowledge tied to physical locations (&#x27;assets&#x27;) where they don&#x27;t have guaranteed network connectivity. Examples include offshore wind farms, power distribution networks, railway infrastructure,  mountain rescue, and military uses.<p>For this demo I query the iPhone app in Airplane mode: &quot;A hiker near Tuolumne Meadows has a dislocated shoulder. What is the reduction protocol, and where is the Tuolumne Meadows Ranger Station?&quot;<p>HOW IT WORKS<p>The core is a portable C++ engine running vector search across an encrypted, on-device SQLite database. The database contains the user&#x27;s proprietary manuals, chunked and converted into embeddings. The search finds relevant chunks and uses them as a prompt for a local LLM (Gemma). This process is RAG (Retrieval-Augmented Generation).<p>Spatial integration is provided by connecting assets in the map to a table in the database and using the table to find any assets referred to in queries and pan the map to them.<p>Everything runs on-device with no API keys or cloud dependencies. The demo app is written in Swift and uses the CartoType framework, which provides a wrapper over the underlying core API giving asynchronous and synchronous access to the AI functions.<p>DATA PREPARATION<p>To get up and running, the user creates a map containing the assets, using CartoType&#x27;s <i>makemap</i> tool, then feeds the map and their proprietary documentation into CartoType&#x27;s <i>makedata</i> tool, which writes the encrypted SQLite database to be stored on the device. The map, which may also be encrypted, is also stored on the device. The CartoType library, when running the Field Assistant&#x27;s RAG system, also provides full map rendering, location searching, routing and geocoding as it always has done.", "5": "2026-08-22T18:05:09.104952"}
{"0": 38, "1": "hackernews", "2": "https://github.com/onecli/onecli", "3": "Launch HN: OneCLI (YC S26) \u2013 OSS sandboxed agent harness for teams", "4": "Launch HN: OneCLI (YC S26) \u2013 OSS sandboxed agent harness for teams. Hi HN, Jonathan &amp; Guy here from OneCLI, an agent harness built for teams, giving every employee a secured, sandboxed personal agent.<p>Here\u2019s what you can do with it:<p>1. get a sandboxed agent, with all the OneCLI capabilities in place like connect your GitHub account, Gmail, Notion, or Dropbox simply from the chat.<p>2. deterministic human in the loop approval in the chat itself for things that you need 100% control like sending an email or deleting the Linear ticket.<p>3. manage team policy in one place, enforced across every agent in the workspace<p>4. enjoy global connections at the team level, like shared LLM keys or service accounts<p>Here\u2019s a demo: <a href=\"https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=dlW-44ntpbE\" rel=\"nofollow\">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=dlW-44ntpbE</a><p>We started working on this by accident, even though our careers were in the security space. We were working on a devtool called ChartDB, an open-source DB tool. When OpenClaw took off back in January, we started using it to orchestrate agents on top of ChartDB. We quickly understood there is a big issue around auth. Agents need credentials to do real work, but to give them those secrets would not be the best idea. They keep them in their memory and also write them down to local files and their sessions as plain text. And we knew that agents can easily be fooled into giving up those API keys&#x2F;secrets. So we needed some way to control the agent and stop prompt injections from tricking it into using its services for an attacker&#x27;s benefit.<p>We created OneCLI that started as a vault for AI Agents built in Rust.<p>We found out that most of our demand for OneCLI came from autonomous agents like Hermes, OpenClaw and NanoClaw for individuals and teams.<p>Users looked for useful agents that do things for the person who runs them with two missing parts: 1) managing secrets and permissions. 2) and for teams - multiplayer management.<p>We decided to pivot and provide the agent itself as a harness for teams, to give each employee an agent. We saw that teams had to deal with setting up their own harness again and again, and basically as we already had the vault as a gateway. We got the idea to provide the missing piece of the agent management out of the box and open source it (Apache-2.0, with a small enterprise exception).<p>We&#x27;re open source first - the entire platform, not just a small portion of it like other agents, so companies can actually see the code, evaluate it, and trust it instead of taking our word for it. They run it isolated, in their own environment, fully under their control, at production quality, not a locked black box hosted somewhere else. That means the safety isn&#x27;t just a promise, it&#x27;s something they can verify themselves. Combined with real autonomy and least-privilege access, that&#x27;s what makes it something a company can fully own and trust, not just adopt.<p>We also approach this from a company perspective rather than an individual one. Our solution manages agents on behalf of each employee, wrapped in deterministic guardrails that company admins configure through centralized policies.<p>For the agent engine itself we\u2019re using jcode which is the core of the agent-loop. We found out that it improves the experience and makes the agent smarter and faster.<p>Here\u2019s how it works:<p>It runs on infra you control. Fully open-source, self-host or cloud in minutes.<p>The agent never holds a real secret. It gets a placeholder. The real credential is injected at the gateway, per request, after the call is authorized. It never enters the agent&#x27;s context, memory, or logs.<p>Enforcement outside the model. Prompts are suggestions. Policies defined by the org admin run at the network layer, outside the agent and the LLM. Block endpoints, rate limit per agent, require approval, scope per employee. The gateway decides. The agent can&#x27;t bypass it.<p>Isolated VM per agent. Own memory, own keys, own permissions. Blast radius is one agent.<p>Speed of the Harness: Rust engine under the agent loop.<p>Full identity trail. Every agent is bound to an employee. Every call logged with who it acted for and which policy allowed it.<p>Some things people are doing with the platform include:<p>- Managing their company life cycle entirely from the sales calls, to the product side automatically open tickets to the engineering teams, that would kick the development agents to deliver and ship to production.<p>- Operational side, like automatically hygiene the CRM after calls, sourcing leads, book meetings and manage follow ups emails.<p>- Some of our customers also doing their entire grocery shopping using those agents and send them to take care of their chores like ordering things online.<p>About the team: Both founders come from cybersecurity backgrounds. Jonathan spent years at Axis Security building zero trust network acces", "5": "2026-08-22T18:05:09.114612"}
{"0": 39, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49360345", "3": "YC is betting on robotics and physical AI startups", "4": "YC is betting on robotics and physical AI startups. Went through 208 startups from YC&#x27;s Winter and Summer 2026 batches to see what they&#x27;re actually funding this year.<p>Robotics was the biggest surprise. There are 32 robotics&#x2F;physical AI companies, exactly as many as AI infrastructure companies. A lot of them are working on pretty specific problems: autonomous factories, CNC machining, warehouse work, waste management, solar construction, nuclear inspection, etc.<p>Some other numbers that stood out:<p>145 of the 208 companies are tagged AI.<p>24 are building developer&#x2F;agent tools. A lot of these aren&#x27;t really about helping you write code anymore. They&#x27;re about managing, monitoring, and debugging coding agents.<p>23 are fintech, 19 healthcare&#x2F;bio.<p>Only 5 are consumer companies. Only 3 are crypto&#x2F;Web3, despite YC explicitly asking for startups in both areas.<p>The teams are tiny too. The median company has just 2 people, and 146 of the 208 have 3 people or fewer.<p>And YC is still very SF: 148 of the 208 companies are based there.", "5": "2026-08-22T18:05:09.121825"}
{"0": 40, "1": "hackernews", "2": "https://maritime.sh", "3": "Show HN: Maritime, a platform for running AI agents for $1 a month", "4": "Show HN: Maritime, a platform for running AI agents for $1 a month. Hi HN, my name is Maria, and I\u2019m a co-founder of Maritime. We started Maritime at MIT to build infrastructure for companies that need to run thousands of isolated AI agents for their customers.<p>Imagine you set up an agent like OpenClaw, or a personal assistant agent with a custom framework, and want to give a separate version of it to every customer&#x2F;friend. Each customer needs their own agent running in an isolated microVM, with persistent state, secrets, triggers, and sleep&#x2F;wake behavior.<p>Building such scalable and secure infra will take you months and will cost hundreds of thousands. Also, you still can&#x27;t vibe code the infra well.<p>So Maritime handles that infrastructure for you.<p>We charge $1 per agent per month, so running 100 isolated agents costs $100 a month.<p>Maritime is designed for companies running thousands of agents, but starting today, any developer can run three agents for free forever.<p>You can use our templates to spin up OpenClaw, Hermes, and the DeepSeek agent and keep all three for free. You can also deploy custom agents through our CLI or SDK and use Maritime as the infra provider for your stratup.<p>You can try it at <a href=\"https:&#x2F;&#x2F;maritime.sh\" rel=\"nofollow\">https:&#x2F;&#x2F;maritime.sh</a>. We\u2019d really appreciate your feedback, especially on the developer experience", "5": "2026-08-22T18:05:09.130434"}
{"0": 41, "1": "hackernews", "2": "https://news.crunchbase.com/venture/physical-ai-funding-startups-robotics-aerospace-h1-2026/", "3": "VCs Pour Billions into Physical AI as the Next Wave of AI Investing Takes Shape", "4": "VCs Pour Billions into Physical AI as the Next Wave of AI Investing Takes Shape. ", "5": "2026-08-22T18:05:09.136695"}
{"0": 42, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49367763", "3": "Ask HN: We have company runway but not founder time. When do you go full-time?", "4": "Ask HN: We have company runway but not founder time. When do you go full-time?. Three first-time founders in Canada, all of us between 45 and 55: business&#x2F;product, marketing, and me on the technical side. We&#x27;re building an AI tool for creative work. About thirty testers have been on a pre-production build for a couple of months. Feedback is good, and importantly is not from friends, but actual target users who we sourced through our networks, which I&#x27;m treating as encouraging signal on product-market fit.<p>We&#x27;re bootstrapped with CAD $25k initial commit and an additional up to $100k available, though we&#x27;d rather earn the right to spend each additional dollar than commit it up front. All three of us still have day jobs, cleared with our employers. At current near-zero usage we could run the service for ~18 months without revenue \u2014 that number obviously falls as real users arrive, which is part of the problem.<p>So our constraint isn&#x27;t the usual one. The company has runway; founder time doesn&#x27;t. I built the platform, I&#x27;m the one who has to go full-time first, and I&#x27;m also the highest-paid person on the team with family obligations that rule out a sudden income cut.<p>For those who started companies in their 40s or 50s: when did you know it was time to leave the day job? Did you wait for revenue to cover you, cut to part-time, build personal runway first, or raise specifically because founder availability had become the bottleneck?<p>Not naming the product \u2014 I am looking for advice, not a launch thread.", "5": "2026-08-22T18:05:10.384361"}
{"0": 43, "1": "hackernews", "2": "https://www.cnbc.com/2026/08/14/open-ai-ipo-red-flag.html", "3": "OpenAI talent exodus raises 'huge red flag' ahead of IPO", "4": "OpenAI talent exodus raises 'huge red flag' ahead of IPO. ", "5": "2026-08-22T18:05:10.391210"}
{"0": 45, "1": "hackernews", "2": "https://techcrunch.com/2026/08/12/ai-coding-startup-cognition-reportedly-already-in-talks-to-raise-at-40b-valuation/", "3": "AI coding startup Cognition in talks to raise at $40B valuation", "4": "AI coding startup Cognition in talks to raise at $40B valuation. ", "5": "2026-08-22T18:05:10.403279"}
{"0": 46, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49276601", "3": "Show HN: A better, free client for iMessage power-users", "4": "Show HN: A better, free client for iMessage power-users. TLDR: Your iMessage inbox is a mess. Go to attention dot righthand dot ai.<p>it&#x27;s a free app for mac. It&#x27;s like Superhuman for iMessage.\n___<p>You text your friends and family in iMessage.  You text your colleagues in iMessage. If you run a business, you likely text prospects and leads in iMessage. I text our customers in iMessage - I want them to know they&#x27;ve got the fastest way to reach me. In 2026 you even text your AI agents in iMessage (Linq raises $20M to enable AI assistants to live in iMessage, Apple approves Poke as the first AI agent on its Messages for Business platform).<p>The point is: your iMessage inbox is overloaded. Even more than email, it&#x27;s become the inbox where everything important happens. But it&#x27;s not built for everything! It&#x27;s basically optimized for friends &amp; family groupchats.<p>Being an iffy texter is OK (debatable) with friends. But the founders and investors we talk to routinely describe this niche pain of co\u00f6rdinating business in iMessage. So we distilled the lessons from Superhuman, viz., the email experience should be fast af and beautiful and shortcut-laden, and we applied them directly to the inboxes that are hard to use - starting with iMessage.<p>Attention is the first app that allows you to become a power-user of iMessage.<p>I&#x27;m finally able to label and divide my iMessages into different folders. I can set reminders and follow-ups with a keystroke.<p>We are obsessed with the potential value of an agent who triages every message before you even see it. We love the concept of every new message arriving with pre-work done and a human-in-the-loop Approval Card curated by the agent with 3 differentiated next steps. It ships with an MCP - just point your favorite flavor of coding agent at it.<p>As far as the actual app: Your data is yours, fully local in LanceDB + SQLite &amp; never leaves your machine unless you turn on the AI agent, at which point data just goes to whichever AI provider you chose.<p>For now, just know that Attention ships with AI features &quot;off&quot; by default and it will always be free to use.<p>attention dot righthand dot ai<p>this post was written without the use of language models by me, Joseph. would love to chat and answer any questions in the comments", "5": "2026-08-22T18:05:10.411750"}
{"0": 47, "1": "hackernews", "2": "https://sequoiacap.com/article/partnering-with-preview-lights-inference-action/", "3": "Preview, a production platform for AI video, raises $12M seed led by Sequoia", "4": "Preview, a production platform for AI video, raises $12M seed led by Sequoia. ", "5": "2026-08-22T18:05:10.417925"}
{"0": 48, "1": "hackernews", "2": "https://www.coderabbit.ai/blog/introducing-agentic-change-management", "3": "CodeRabbit raises a $143M Series C at a $1.5B valuation", "4": "CodeRabbit raises a $143M Series C at a $1.5B valuation. ", "5": "2026-08-22T18:05:10.424777"}
{"0": 49, "1": "hackernews", "2": "https://www.bloomberg.com/news/newsletters/2026-08-11/sequoia-raises-its-bets-on-ai-startups", "3": "Sequoia Raises Its Comfort with Risk in AI Bets", "4": "Sequoia Raises Its Comfort with Risk in AI Bets. ", "5": "2026-08-22T18:05:10.428901"}
{"0": 50, "1": "hackernews", "2": "https://www.morningstar.com/news/business-wire/20260811845258/river-ai-raises-11b-led-by-general-catalyst-and-amp-pbc-to-build-open-ai-stack", "3": "River AI Raises $1.1B Led by General Catalyst and AMP PBC to Build Open AI Stack", "4": "River AI Raises $1.1B Led by General Catalyst and AMP PBC to Build Open AI Stack. ", "5": "2026-08-22T18:05:10.433576"}
{"0": 51, "1": "hackernews", "2": "https://www.reuters.com/technology/wall-street-giants-partner-with-nvidia-500-billion-ai-financing-deal-ft-reports-2026-08-10/", "3": "Nvidia partners with Wall Street giants to raise $500B for AI buildout", "4": "Nvidia partners with Wall Street giants to raise $500B for AI buildout. ", "5": "2026-08-22T18:05:10.439214"}
{"0": 52, "1": "hackernews", "2": "https://www.the-substrate.net/p/why-cant-chinese-ai-companies-raise", "3": "Why can't Chinese AI companies raise more money?", "4": "Why can't Chinese AI companies raise more money?. ", "5": "2026-08-22T18:05:10.444108"}
{"0": 53, "1": "hackernews", "2": "https://www.konsulteer.com/article/berlin-s-telli-raises-15m-to-expand-enterprise-ai-agents", "3": "Berlin's Telli Raises $15M to Expand Enterprise AI Agents", "4": "Berlin's Telli Raises $15M to Expand Enterprise AI Agents. ", "5": "2026-08-22T18:05:10.450025"}
{"0": 54, "1": "hackernews", "2": "https://github.com/berwinsingh/oldhand", "3": "Show HN: OldHand A Claude/Codex plugin to verify the development flow end-to-end", "4": "Show HN: OldHand A Claude/Codex plugin to verify the development flow end-to-end. I built this clod code or codec skill because I just love using these AI agents to write code. It&#x27;s just so convenient to do. But when it comes to working in a corporate environment, I have to ultimately follow a Jira ticket or a Asana ticket.<p>Now, when I&#x27;m following those tickets, I am able to do the work, but it&#x27;s not always up to the mark because after all I found AI to hallucinate, Unnecessary code is being written, even though a function for that is already there and can be reused, no proper testing apart from unit test being done so when the QA started testing it they raised a astronomical amount of bugs.<p>So, with Claude and Codex becoming very capable especially with their plugins&#x2F;connectors and computer use ability, I build OldHand that allows for end-to-end ticket development, planning, coding, reusability (combined with Ponytail), QA testing with browser&#x2F;computer use.<p>I would love to get people&#x27;s feedback as it has been helping me a lot with my day-to-day work, but I would like to improve this even further and every feedback helps.", "5": "2026-08-22T18:05:10.458182"}
{"0": 55, "1": "hackernews", "2": "https://playdowntime.com/", "3": "Show HN: My journey into game development with AI", "4": "Show HN: My journey into game development with AI. I had a development background initially, moved out to a different field 20+ years ago, but often have a nostalgia with periodic attempts to create something. Wanted to share the last one, started 2 months ago as part of my upskilling with AI. Choice what to develop was easy - spent lots of time commuting to the office and wanted a simple game playable offline, without ads, within 5-10 minutes. I have never developed mobile games and this was an point of the experiment - how far I can reach? :)<p>Stage 0 (full control &amp; no trust) - I started with VS Code + Android Studio. Yes, Claude created something, however compilation and error correction was an old-style pain. Initially googled, then asked AI &quot;check and correct the error&quot;. Then I realised that I don&#x27;t need neither VS Code nor Android Studio. When I need to review a code, i can do it in project folder directly.<p>Stage 1 (building the trust) - moved to Claude Code CLI. Jointly with Claude we defined the processes and the team structure (agents responsible for core framework, games, QA tests, and the tech lead performing the code review), and then Claude set up the environment (bye-bye VS Code). At this stage, I still reviewed some code but started to build trust into overriding reviews by tech lead.<p>Stage 2 (direction and supervision of the AI team) - I realised that it&#x27;s more valuable to have a proper planning to define the requirements (concept, high-level architecture and design), and then allow agents to develop the game, add automated QA tests, and then tech lead to do a code review.<p>Some rules on the road and helpful processes that we defined on this journey:\n- Key architectural decisions are fixed during the initial planning. AI is not allowed to circumvent them and must ask my approval, and in most cases, we found better ways within the initial boundaries. The same applies to the core app framework that was frozen after the initial implementation and QA.\n- Hard module ownership - prevents agents to make quiet changes to each other&#x27;s code. Technical lead still flags such cases during review, but these are rare exceptions now. \n- UI design - Claude can do UI, but it&#x27;s not the strongest area, and initially it was a pain... We established a workflow when all UI changes are first shown on local web server. It still takes multiple iterations to get to appropriate level, but much easier now. \n- Device testing matrix - it allows to see how it looks like on different screens (resolution, density, aspect). We ended up in 12 types from low-end to high-end phones and tablets. One script runs app on emulator and takes screenshots. Another script identifies issues in app screens. The last presents all screenshots (by devices, layouts based on portrait&#x2F;landscape, touch&#x2F;buttons, dark&#x2F;light theme) on local web server. Once all ok, the screenshots are promoted as golden and used in later QA tests.\n- Triage flag - helpful when I&#x27;m away and have some ideas or thoughts. I simply raise an issue on Github mobile, add a triage flag, and Claude picks it in our next session.\n- Detailed test plans for manual testing on live devices.\n- For marketing materials, AI&#x27;s focus is shifted towards low-level details. It&#x27;s helpful to propose some points or do a sense check of human-written draft. But if you want a story with a correct focus - do it manually...<p>Key controls in the process are:\n- peer review by tech lead agent for all changes; \n- automated QA tests and UI tests based on the device matrix; \n- standardised release process (validation of QA tests, final testing of AAB release, updates to Play Store listing and website, and some other steps);\n- Play Store upload is still a manual step, done by me when happy with all release process checks and steps.<p>And a little about the application itself - it&#x27;s free, offline, currently contains 2 games (easy classic games where you don&#x27;t need to learn how to play), and more games are coming :) Web-page: <a href=\"https:&#x2F;&#x2F;playdowntime.com\" rel=\"nofollow\">https:&#x2F;&#x2F;playdowntime.com</a>", "5": "2026-08-22T18:05:10.462309"}
{"0": 56, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49156487", "3": "Can a robotics startup survive by pivoting from hardware to data scraping?", "4": "Can a robotics startup survive by pivoting from hardware to data scraping?. According to public posts from one of their engineers, they&#x27;re now:<p>\u00b7 Scraping 1M+ hours of POV video from the internet\n\u00b7 Burning significant GPU hours while models only work for basic tasks\n\u00b7 Down to &lt;10 people after laying off ~50% of the team<p>This raises a broader question:<p>\u00b7 Is scraping public video a viable long-term data strategy, or a dead end?\n\u00b7 Are there examples of hardware companies that successfully made this pivot?<p>They originally built Eggie, a multipurpose humanoid robot, but Tangible has since pivoted to building an AI foundation model for robotics.", "5": "2026-08-22T18:05:10.468906"}
{"0": 57, "1": "hackernews", "2": "https://www.theguardian.com/technology/2026/aug/15/uk-ireland-booksellers-suspect-ai-companies-bulk-orders-data-acquisition", "3": "UK & Ireland Secondhand Booksellers Suspect AI Firms Behind Strange Bulk Orders", "4": "UK & Ireland Secondhand Booksellers Suspect AI Firms Behind Strange Bulk Orders. ", "5": "2026-08-22T18:05:11.901867"}
{"0": 58, "1": "hackernews", "2": "https://medium.com/@glennlenormand/stripe-didnt-pay-7-billion-for-openrouter-it-bought-the-tollbooth-of-the-ai-economy-5849f9891488", "3": "Why Stripe's $7B OpenRouter Acquisition Is About Owning the AI Routing Layer", "4": "Why Stripe's $7B OpenRouter Acquisition Is About Owning the AI Routing Layer. ", "5": "2026-08-22T18:05:11.907526"}
{"0": 59, "1": "hackernews", "2": "https://news.bloomberglaw.com/mergers-and-acquisitions/stripe-nears-deal-to-buy-ai-firm-openrouter-for-over-7-billion", "3": "Stripe Nears Deal to Buy AI Firm OpenRouter for over $7B", "4": "Stripe Nears Deal to Buy AI Firm OpenRouter for over $7B. ", "5": "2026-08-22T18:05:11.912665"}
{"0": 62, "1": "hackernews", "2": "https://archer.com/news/archer-to-shape-physical-ai-future-of-aerospace-and-defense-with-acquisition", "3": "Archer to Acquire Wisk, Insitu, and SkyGrid from Boeing", "4": "Archer to Acquire Wisk, Insitu, and SkyGrid from Boeing. ", "5": "2026-08-22T18:05:11.932123"}
{"0": 63, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49178393", "3": "Do You Think OpenAI Is Apple Circa the 1980s?", "4": "Do You Think OpenAI Is Apple Circa the 1980s?. OpenAI is a deeply mismanaged company. Most recently, a blog post that feels like it has got to have dozens of PR violations, responding to the Apple lawsuit.<p>More broadly, OpenAI\u2019s main problem is that it does not have a real competitive advantage. No real models do; the real differentiation is price; they will become commodities. The need to move upwards in the market in enterprise software is difficult since the current stack should remain the same, since sticking with Salesforce plus its new AI features is much simpler than the cost of switching to an AI-first startup version, which won\u2019t be around 3 to 5 years from now. So consumer hardware and trying to become the next Apple could work, except you would likely have to focus on people not in the Apple ecosystem because of the double-sided lock-in Apple has.<p>The opposite dynamic works in consumer and mobile, really. Apple with AI is a much worse experience than something AI-first built from the ground up. Especially considering the consumer market does not have the switching costs enterprise does. So clearly OpenAI has some kind of opportunity to be a disruptor, just not to Apple, likely a large chunk of Android users, based on what their new device is.<p>So while at first it may look like OpenAI is doing too much, it is a desperate attempt to create some kind of real, defensible business, and show some proof of such before they IPO. If they do not do so, public market investors will cut their throats, and they will see a valuation drop like no other, regardless of where we are in the capital cycle. If they delay IPO plans, it will become clear they are in shambles. They are far behind in the enterprise game, which is short-term anyway.<p>Their best bet is hardware; credit to Altman and co for realizing this, although Apple is a roadblock, which I hope, for the sake of competition, is removed. Hardware is interesting, and perhaps somewhat disruptable towards Apple since OpenAI would be vertically integrated in a way Apple cannot since they do not have their own models. The difficulty is whether the burn OpenAI is spending on multiple fronts can be sustained long enough to see some promise. No doubt the consumer app is big, in terms of users, yet there is not enough activity to create a large ad business. The reason for this is Google and AI overviews, which are a far better user experience. So perhaps consumers can be an expensive customer acquisition cost to jumpstart the hardware business in some way? Regardless, OpenAI does have the right strategy in terms of attempting to create something defensible and long-term.<p>Now, Anthropic having a more coherent strategy in the short term of focusing on enterprise from the start is better; it is still not defensible. Perhaps because of the fact that the cost of paying Anthropic is far too great and outweighs any form of customer captivity. Plus, an enterprise gets maximum leverage by training a model, ideally a cheap one, on its own data so it can get insights tailored to the enterprise specifically, something Anthropic cannot deliver. Anthropic\u2019s fall will likely be in line with the broader capital cycle.<p>All in all, both companies as it stands today are massively overvalued, even Anthropic with its sky-high revenue, which is not real considering the capital cycle and short-term interest&#x2F;desperation of enterprises, especially when there is a less expensive, far more valuable way to implement AI through using open-source models and training them on your data. Application layer companies such as BaseTen and OpenRouter should benefit from building on top of these new commodities. OpenAI, as of now, has the only long-term viable strategy, which has a lot of execution risk, yet excites me about the future of OpenAI.", "5": "2026-08-22T18:05:11.938072"}
{"0": 64, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49150707", "3": "Ask HN: What Is Stripe Today?", "4": "Ask HN: What Is Stripe Today?. I think Stripe\u2019s internal strategy is a consolidation of products within its sector. Similar across other industries, such as cybersecurity, where I think Cloudflare is attempting to do the same. The broader idea is clearly enterprise consolidation. I think now is the correct period in the technology cycle for such consolidation.<p>The influx of many enterprise companies was due to the advent from on premise computing&#x2F;storage to the cloud, lowering the barrier to creating new companies. Hence, it was somewhat possible to create \u201cNiche\u201d enterprise businesses, such as a Dropbox or an Intercom. These businesses were able to create sustainable recurring profits because of a business model boom and customer captivity in terms of enterprise switching costs. Hence, none of these businesses are disappearing and should remain stable companies for a long time. There is a broader move to add company-adjacent products to further entrench customers and somewhat diversify from a core business.<p>As costs rise for enterprises via AI spending, there is a need to cut costs and not pay for 20 different niche enterprise solutions. So there is a market need for a more bundled experience. Stripe has launched several products around the core payments business to help with accounting, taxes, and incorporation. Cloudflare is doing something similar by creating a suite of cybersecurity products, all under one roof, VPN\u2019s, boxes, CDN\u2019s.<p>This broader repeated concept of new technology leads to fragmented markets, then slower growth, and a need to expand the core business to continue to grow and expand their respective businesses.<p>So this is further strengthened by the recent acquisition efforts Stripe has made in PayPal and OpenRouter. PayPal is interesting because it gives Stripe a powerful brand and acceptability in payments on the internet. It also gives Stripe an entry into the consumer market and p2p payments. OpenRouter is more interesting because of its potential position in the AI value chain. There still does not seem to be any kind of economic barrier protecting frontier models from an inevitable commoditization. OpenRouter acts as an aggregator of sorts, and could be the premier application in the AI value chain.<p>While both are still under negotiation, it highlights the broader strategy of building a suite of products. This creates a much more valuable business because the likelihood of growing and defensible cash flows increases. You can switch, perhaps all your files on your Dropbox to a GDrive, difficult but not unreasonable. You likely cannot switch your payments, your taxes, your accounting, and additional products.<p>Overall consolidation is just a natural phase in the industry, and perhaps the most value accruable to shareholders. It is the strategy of investing based on capital cycles and an interesting development in the current state of enterprise SaaS. The current new wave of companies building AI first is likely to not stick around. New technologies are almost always better bundled in the enterprise world. Figma is a much more enhanced product with AI, not irrelevant. Though to build defensibility, you cannot just be a one-trick pony; surely companies that are really just a feature will struggle as this wave of fragmented SaaS comes to an end.", "5": "2026-08-22T18:05:11.945713"}
{"0": 65, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49129778", "3": "Is Stripe the Internet?", "4": "Is Stripe the Internet?. I think Stripe\u2019s internal strategy is a consolidation of products within its sector. Similar across other industries, such as cybersecurity, where I think Cloudflare is attempting to do the same. The broader idea is clearly enterprise consolidation. I think now is the correct period in the technology cycle for such consolidation.<p>The influx of many enterprise companies was due to the advent from on premise computing&#x2F;storage to the cloud, lowering the barrier to creating new companies. Hence, it was somewhat possible to create \u201cNiche\u201d enterprise businesses, such as a Dropbox or an Intercom. These businesses were able to create sustainable recurring profits because of a business model boom and customer captivity in terms of enterprise switching costs. Hence, none of these businesses are disappearing and should remain stable companies for a long time. There is a broader move to add company-adjacent products to further entrench customers and somewhat diversify from a core business.<p>As costs rise for enterprises via AI spending, there is a need to cut costs and not pay for 20 different niche enterprise solutions. So there is a market need for a more bundled experience. Stripe has launched several products around the core payments business to help with accounting, taxes, and incorporation. Cloudflare is doing something similar by creating a suite of cybersecurity products, all under one roof, VPN\u2019s, boxes, CDN\u2019s.<p>This broader repeated concept of new technology leads to fragmented markets, then slower growth, and a need to expand the core business to continue to grow and expand their respective businesses.<p>So this is further strengthened by the recent acquisition efforts Stripe has made in PayPal and OpenRouter. PayPal is interesting because it gives Stripe a powerful brand and acceptability in payments on the internet. It also gives Stripe an entry into the consumer market and p2p payments. OpenRouter is more interesting because of its potential position in the AI value chain. There still does not seem to be any kind of economic barrier protecting frontier models from an inevitable commoditization. OpenRouter acts as an aggregator of sorts, and could be the premier application in the AI value chain.<p>While both are still under negotiation, it highlights the broader strategy of building a suite of products. This creates a much more valuable business because the likelihood of growing and defensible cash flows increases. You can switch, perhaps all your files on your Dropbox to a GDrive, difficult but not unreasonable. You likely cannot switch your payments, your taxes, your accounting, and additional products.<p>Overall consolidation is just a natural phase in the industry, and perhaps the most value accruable to shareholders. It is the strategy of investing based on capital cycles and an interesting development in the current state of enterprise SaaS. The current new wave of companies building AI first is likely to not stick around. New technologies are almost always better bundled in the enterprise world. Figma is a much more enhanced product with AI, not irrelevant. Though to build defensibility, you cannot just be a one-trick pony; surely companies that are really just a feature will struggle as this wave of fragmented SaaS comes to an end", "5": "2026-08-22T18:05:11.952908"}
{"0": 66, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49116817", "3": "What's Working for Adoption and Distribution of AI/SaaS Products in 2026?", "4": "What's Working for Adoption and Distribution of AI/SaaS Products in 2026?. Building is (supposedly) easy and cheap. \nMedia showcases the AI hits that grow faster than any company in history. \nYouTube &quot;experts&quot; are touting AI&#x2F;SaaS tools that grew to $1m ARR within 3 months.<p>But, that&#x27;s only a small slice of the world. We have been building useful products that drive real results and our customers like...but, without big funding and limited people, we can&#x27;t seem to rise above the noise to get any traction.<p>So, for those without millions to spend (lose) on customer acquisition...what is working in 2026 to get a few users per day to give new tools a shot?", "5": "2026-08-22T18:05:11.959736"}
{"0": 67, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49107855", "3": "Ventora Expands Its AI Business Builder to Help Solo Founders", "4": "Ventora Expands Its AI Business Builder to Help Solo Founders. Building software is becoming easy.\nBuilding a business is not.<p>AI coding tools have dramatically lowered the barrier to creating SaaS products, AI applications, online services, marketplaces, and e-commerce stores. Entrepreneurs can now generate working products in hours instead of spending months writing code.<p>But for most founders, development is no longer the bottleneck.<p>Launching a successful business still requires validating demand, researching competitors, defining positioning, creating marketing assets, acquiring customers, optimizing advertising, and continuously improving performance. These activities often require an entire startup team with expertise across multiple disciplines.<p>Ventora is expanding its AI Business Builder platform around a different idea: entrepreneurs shouldn&#x27;t need to assemble that team before testing an opportunity.<p>Instead of focusing solely on product generation, the platform combines specialized AI agents that support business analysis, market validation, product creation, launch preparation, marketing execution, and customer acquisition within a unified workflow.<p>The goal is to help entrepreneurs spend less time coordinating developers, marketers, designers, and analysts, and more time identifying valuable opportunities and making strategic decisions.<p>According to the company, this approach allows founders to move from an idea to a functioning online business significantly faster than traditional startup workflows while reducing the operational complexity that has historically prevented many people from launching companies.<p>The platform is designed for entrepreneurs building SaaS products, AI tools, subscription businesses, digital marketplaces, online services, e-commerce stores, and other internet businesses.<p>Rather than replacing founders, Ventora aims to replace much of the repetitive operational work traditionally handled by multiple specialists, allowing a single entrepreneur to accomplish what previously required an entire early-stage team.<p>As AI continues to evolve beyond code generation, platforms are beginning to automate larger portions of business creation itself. Ventora believes this represents the next stage of AI adoption: not simply helping people build software, but helping them build businesses.", "5": "2026-08-22T18:05:11.966492"}
{"0": 68, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49048352", "3": "The Cost of Seamlessness", "4": "The Cost of Seamlessness. The Hidden Cost of Frictionless Technology\nTaotuner \u2014 2026<p>I recently asked an AI assistant for restaurant recommendations. It listed five places, complete with ratings, distance, and popular dishes. I picked one, went, had a fine meal. Later, when a friend asked what I&#x27;d eaten, I couldn&#x27;t remember the name of the restaurant.<p>The experience wasn&#x27;t bad. It was frictionless. And it left almost no trace.<p>We&#x27;ve built tools that eliminate the gap between want and satisfaction, question and answer, curiosity and resolution. We call it seamlessness. We&#x27;ve assumed it&#x27;s progress. But the gap wasn&#x27;t inefficiency \u2014 it was where memory, judgment, and understanding used to form.<p>What if removing it is making us less capable?<p>The Pattern\nThis isn&#x27;t just about restaurant recommendations. The same dynamic shows up across AI, therapy, cities, and education.<p>Take AI assistants. The industry evaluates them on speed, accuracy, and satisfaction. Those metrics capture whether the tool works. They don&#x27;t capture what the tool does to the person using it.<p>When an AI answers before you finish your question, it saves time. It also interrupts the process of figuring out what you actually wanted to ask. When it completes your sentence, it&#x27;s efficient. It also prevents you from discovering what you were trying to say.<p>Researchers at Harvard and MIT found something unsettling: when radiologists used AI assistants, their diagnostic accuracy improved. But when the AI wasn&#x27;t available, their unaided performance had declined. They had become dependent \u2014 faster with the tool, worse without it.<p>Studies on AI coding assistants show the same pattern. Developers complete tasks faster. But they also write more buggy code than those who did the same tasks unaided \u2014 and struggle to fix bugs without the AI. They&#x27;ve outsourced the reasoning, not just the typing.<p>Therapy offers a counterexample. A good therapist doesn&#x27;t rush to make you feel better. They sustain the conditions under which you can do the work \u2014 and that requires a specific kind of silence. The pause where something might form.<p>Collapse that silence with reassurance or advice, and you interrupt the process. The distress might drop. But the integration doesn&#x27;t happen.<p>Now consider what happens when AI enters that space. A person in distress at 2am opens a chatbot. The distress decreases. But the question that was forming \u2014 the real material for the next session \u2014 was answered before it crystallized. The tension that, held until morning, might have produced insight: resolved, and gone.<p>Not all relief is therapeutic. Most platforms aren&#x27;t designed to know this.<p>Cities show the same pattern. When planners over-optimize \u2014 separating uses, eliminating ambiguity, programming every space \u2014 the city becomes efficient and sterile. It stops generating the unplanned encounters where culture and solidarity form.<p>Jane Jacobs described this decades ago. She argued that cities are &quot;problems in organized complexity&quot; \u2014 they deal with many interrelated factors that form an organic whole. The order of a well-adapted city emerges from human action, not human design. Mixed uses, short blocks, aged buildings, density \u2014 these create the conditions for unplanned encounters.<p>The modernist housing project Pruitt-Igoe was designed with efficiency in mind: clean lines, separated functions, clear circulation. Within two decades, it was demolished. The design eliminated the productive tension \u2014 the messy mix of uses, the unexpected intersections \u2014 that makes a city livable.<p>There&#x27;s a deeper layer here. Yuk Hui argues that technology is never neutral. Every city embodies a cosmology \u2014 a way of understanding the world and how we should inhabit it. Modern urbanism embodies a cosmology of control: every use anticipated, every space programmed. It treats the city as a machine to be optimized.<p>Other cosmologies are possible. The traditional Chinese garden embodies a different logic: not control, but relation. It invites participation rather than prescribing use. The alternative to the machine-city isn&#x27;t chaos. It&#x27;s what some call intentional incompleteness: spaces that invite appropriation rather than prescribing it. A plaza where people rearrange furniture. A street that leaves room for negotiation between pedestrians and cars. A library with unprogrammed zones. Not a fixed design. A field.<p>Education is where this hurts most. EdTech companies know that difficulty is not an obstacle to learning \u2014 it&#x27;s the mechanism.<p>The &quot;generation effect&quot;: information is better remembered when you generate it yourself. &quot;Desirable difficulty&quot;: learning improves with effort. Robert Bjork distinguishes performance from learning. Performance is now. Learning is later. Manipulations that speed up apparent acquisition can fail to sustain long-term retention.", "5": "2026-08-22T18:05:11.971719"}
{"0": 69, "1": "hackernews", "2": "https://www.getvertical.ai/blog/garmin-trainingpeaks-acquisition/", "3": "Why Garmin Bought TrainingPeaks", "4": "Why Garmin Bought TrainingPeaks. ", "5": "2026-08-22T18:05:11.977863"}
{"0": 70, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=48996643", "3": "We're (Skyfall AI) acquiring SaaS startups and running them with AI as CEO", "4": "We're (Skyfall AI) acquiring SaaS startups and running them with AI as CEO. Skyfall AI came out of stealth this week and is actively acquiring early-stage SaaS businesses to serve as proving grounds for our AI.<p>Here&#x27;s what we&#x27;re looking for:<p>- Monthly recurring revenue under $20,000\n- Founded within the last 24 months\n- Self-serve or product-led growth (no sales calls required)\n- Customers can sign up and pay without human interaction<p>What we&#x27;re offering:<p>- Up to $1 million per acquisition\n- Our AI takes on the CEO role and operationalizes the business\n- Target: ideally double revenue in the first six months<p>We&#x27;re looking for founders who are willing to take a bet on this approach. If you&#x27;re interested or know someone who is, we&#x27;d like to hear from you (submit your company here\u201d https:&#x2F;&#x2F;skyfall.ai&#x2F;acquisition ).<p>PS: We\u2019re accepting submissions through August 31!", "5": "2026-08-22T18:05:11.983722"}
{"0": 71, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=48981136", "3": "Launch HN: Bloomy (YC S26) \u2013 AI-powered mastery learning for K-12", "4": "Launch HN: Bloomy (YC S26) \u2013 AI-powered mastery learning for K-12. Hi HN, I\u2019m Alex Southmayd, the founder of Bloomy (<a href=\"https:&#x2F;&#x2F;bloomylearning.com\">https:&#x2F;&#x2F;bloomylearning.com</a>) \u2013 an AI-powered mastery-learning platform for K-12 students. Bloomy provides students with an AI tutor alongside adaptive curriculum (right now Math, English Language Arts, and Writing).<p>How it works: we diagnose students\u2019 skill gaps, place them on personalized learning paths, and give them standards-aligned lessons and a Socratic AI tutor that scaffolds their learning without just giving away the answer.<p>The goal is to solve the Bloom 2-sigma problem (<a href=\"https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Bloom%27s_2_sigma_problem\" rel=\"nofollow\">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Bloom%27s_2_sigma_problem</a>) with AI.<p>Short launch video: <a href=\"https:&#x2F;&#x2F;tinyurl.com&#x2F;bloomylearning\" rel=\"nofollow\">https:&#x2F;&#x2F;tinyurl.com&#x2F;bloomylearning</a><p>Longer product demo: <a href=\"https:&#x2F;&#x2F;youtu.be&#x2F;XHvoKt6qMeo\" rel=\"nofollow\">https:&#x2F;&#x2F;youtu.be&#x2F;XHvoKt6qMeo</a><p>Families access for Bloomy: <a href=\"https:&#x2F;&#x2F;bloomylearning.com&#x2F;families\">https:&#x2F;&#x2F;bloomylearning.com&#x2F;families</a><p>I started as a teacher. I taught 7th-grade English and writing with Teach For America, and every day I struggled to deliver differentiated instruction to 30 students with 30 different sets of needs. Some students needed remediation, some needed acceleration, and many needed a tutor sitting next to them helping them reason through the next step.\nBenjamin Bloom\u2019s two-sigma result\u2014that one-on-one tutoring can produce much better outcomes than conventional classroom instruction\u2014always felt intuitively true to me. The hard part was making that kind of attention affordable and available to every child.<p>Then AI changed the cost curve. When I saw schools such as Alpha organize academics around mastery rather than seat time, the model clicked. If you\u2019ve heard of Alpha School, that is directionally the kind of learning model that inspired us. But I kept thinking about the families and schools that already exist: homeschool families, microschools, hybrid schools, and regular classrooms where most children are today.<p>Most students and teachers see learning gaps at the wrong resolution. They get a grade, percentile, benchmark score, or broad standard\u2014not \u201cthis is the next skill this student should learn.\u201d Existing personalized-learning products often feel like digital worksheets: they provide plenty of practice, but not much diagnosis or teaching. Very few have AI tutors providing the core instruction.\nBloomy starts with a diagnostic\u2014we integrate with third-party assessments and provide our own\u2014and creates a learning path for each student. Students work one skill at a time, receive a short lesson, practice at an adaptive difficulty, and only move forward after demonstrating at least 90% mastery. The learning path updates as the student works, based on their performance and our knowledge graph of skill prerequisites (built in collaboration with Learning Commons &#x2F; Chan Zuckerberg Initiative).<p>Each skill has three stages. Base Camp teaches the concept with worked examples. Climb provides guided practice and Socratic support. Summit is an independent ten-question mastery assessment with no hints or AI assistance. Students need to achieve 90% on the Summit to advance. If they struggle too much, they\u2019ll be routed to a different skill better suited for their level.<p>BloomyBot is not a blank chat window but rather a live, interactive, and observant digital tutor. During practice, it receives the active passage or problem, the question, the student\u2019s attempt, an authored explanation, and relevant misconception context. It follows a scaffolded tutoring ladder: first asking what the student tried, then pointing toward the concept, suggesting a strategy, working through one step together, and only providing heavier scaffolding after the student has struggled, adapting to and learning from the student along the way. Students can interrupt it, and we\u2019ve begun to roll out multilingual support for Spanish, French, and a few other more niche languages that customers have asked for.<p>We currently use a variety of Anthropic and OpenAI models for BloomyBot. The tutor is restricted to the current lesson, redirects unrelated questions, limits conversation length, and is unavailable during mastery assessments. The language model does not choose the curriculum or decide whether a student has mastered a skill.<p>That separation is important. A conventionally \u201chelpful\u201d AI response can be a bad tutoring response: if it gives away the answer, the student completes the task but may not learn anything. Our goal is not to build a homework-answering chatbot. It is to put AI inside a structured loop of diagnosis, instruction, practice, feedback, and independent master", "5": "2026-08-22T18:05:11.989555"}
{"0": 72, "1": "hackernews", "2": "https://onlinelibrary.wiley.com/doi/10.1002/aaai.70061", "3": "Six principles for evaluating cognitive capabilities in AI models", "4": "Six principles for evaluating cognitive capabilities in AI models. ", "5": "2026-08-22T18:05:13.247894"}
{"0": 73, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49378057", "3": "Are you good at AI, or just using it?", "4": "Are you good at AI, or just using it?. We\u2019re working on a ladder for individual AI proficiency and would love feedback on both the levels and the definitions.<p>L0 New: brand new to AI, or has not yet used it.<p>L1 Chat: simple prompt-and-response use. Work is serial: ask, wait for an answer, then ask again.<p>L2 Contextual Work: gives AI relevant documents, data, or workspace context so it can work within the actual artifact and produce a more useful result.<p>L3 Orchestrate: coordinates multiple agents or AI roles across independent workstreams, with work that may review, challenge, compare, or build on other work. This is not just for engineering.<p>L4 Automate: creates workflows that are triggered by business events and run without someone sitting at a laptop directing each step.<p>L5 Loop: feeds the output of those workflows back into shared knowledge or a company brain, so future workflows improve over time.<p>A few things I\u2019d love your perspective on:<p>Are these the right levels?\nAre any of the names unclear or overlapping?\nWhat observable behaviors would you use to distinguish one level from the next?\nDoes \u201cloop\u201d make sense as an individual proficiency level, or is it inherently a team or company capability?\nIs there a L6 and if so how would you define it?\nWe\u2019re trying to define these because, in customer conversations, we\u2019ve found that people are not very good at self-evaluating their own AI proficiency. Frequent use often gets mistaken for proficiency. And being low on a ladder like this can feel like admitting you are falling behind, do not fit in, or are less secure in your job, especially for leaders expected to set the pace. We want a more objective, behavior-based way to distinguish the two.", "5": "2026-08-22T18:05:13.257055"}
{"0": 74, "1": "hackernews", "2": "https://techcrunch.com/2026/08/18/etcheds-valuation-doubles-to-21b-in-a-month/", "3": "Etched AI valuation doubles to $21B in a month", "4": "Etched AI valuation doubles to $21B in a month. ", "5": "2026-08-22T18:05:13.263555"}
{"0": 75, "1": "hackernews", "2": "https://github.com/NAEOS-foundation/naeos", "3": "Show HN: Naeos \u2013 an engineering system for AI coding agents", "4": "Show HN: Naeos \u2013 an engineering system for AI coding agents. NAEOS is an open-source engineering system for AI coding agents.It provides architecture, engineering standards, policies, specifications, and validation wworkflow to help agents build software within a consistent engineering context", "5": "2026-08-22T18:05:13.268499"}
{"0": 76, "1": "hackernews", "2": "https://www.cnn.com/2026/08/18/tech/china-unitree-ipo-intl-hnk", "3": "Unitree Robotics IPO: \"Embodied AI\" Valuation Bubble?", "4": "Unitree Robotics IPO: \"Embodied AI\" Valuation Bubble?. ", "5": "2026-08-22T18:05:13.277149"}
{"0": 77, "1": "hackernews", "2": "https://www.reuters.com/technology/ai-chip-startup-etched-valued-21-billion-latest-funding-round-2026-08-18/", "3": "AI chip startup Etched doubles valuation to $21B in under a month", "4": "AI chip startup Etched doubles valuation to $21B in under a month. ", "5": "2026-08-22T18:05:13.286958"}
{"0": 78, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49345378", "3": "Is there room for an AI-powered webmail service built on a single-letter domain?", "4": "Is there room for an AI-powered webmail service built on a single-letter domain?. Hi HN, I\u2019m a solo, non-technical founder working on W, an AI-powered webmail service built on a premium single-letter domain (https:&#x2F;&#x2F;w.xyz).<p>I secured the domain and used AI agents to build out a functional proof-of-concept for the client. To handle infrastructure, I routed the application logic through Mailgun. Internally, the prototype is fully operational \u2014 emails successfully send and receive from @w.xyz handles. W combines a highly optimized interface with built-in AI capabilities like Spam Scanning, Smart Reply and Text Summaries.<p>I currently have a 30+ organic waitlist as I finish the final touches and prepare to move away from the prototype stack.<p>My questions for the community:<p>Concept Validation: Given how crowded the email space is, do you see a viable market for an AI-native premium webmail service, or do you believe the switching cost from Gmail&#x2F;Outlook too high for users?<p>Scaling Roadmap: As a non-technical founder moving away from an AI-generated prototype stack, what are the most critical architectural baselines I must prioritize right now to ensure the system scales smoothly?<p>P.S. If you&#x27;re interested in building a new kind of email platform from the ground up, drop a comment below.", "5": "2026-08-22T18:05:13.292949"}
{"0": 79, "1": "hackernews", "2": "https://www.smh.com.au/technology/sobering-markdown-canva-slashes-valuation-by-10b-as-ai-reality-bites-20260814-p60of9.html", "3": "Canva cuts internal value by $10B amid AI market fears", "4": "Canva cuts internal value by $10B amid AI market fears. ", "5": "2026-08-22T18:05:13.300201"}
{"0": 80, "1": "hackernews", "2": "https://www.msn.com/en-au/money/news/sobering-markdown-canva-slashes-valuation-by-10b-as-ai-reality-bites/ar-AA2a5i6F", "3": "Canva slashes valuation by $10B as AI reality bites", "4": "Canva slashes valuation by $10B as AI reality bites. ", "5": "2026-08-22T18:05:13.307537"}
{"0": 81, "1": "hackernews", "2": "https://www.codewithbullet.com", "3": "Launch HN: Bullet (YC S26) \u2013 A Faster Coding Agent", "4": "Launch HN: Bullet (YC S26) \u2013 A Faster Coding Agent. Hi HN! We\u2019re Adi and Alex, founders of Bullet, a faster coding agent.<p>Bullet started in a senior year dorm. We were fresh out of working at AppLovin and Citadel, and naturally thought we were on a sure path to startup success. We were going to use our skills optimizing stock pricing calculation speeds and agent document context to take over the world. So, Bullet started as an AI hedge fund, a browser-use agent, synthetic financial data (oof), a mobile IDE, and a bunch of other things. We wanted to build something people wanted, but it seemed like everything we built was just terrible, useless, or both.<p>So, we decided to do something completely different, something completely out of the blue, something that no one had ever done before. Solve a problem we actually had.<p>Over the course of six pivots, we suffered. Throughout all of our adventures, one final boss kept getting in our way. Claude Code and his little brother Codex. We were spending hours waiting for coding agents like Claude Code and Codex, and got so frustrated to the point that I downloaded the Claude Code whip. We had spent months of time waiting for six codebases-worth of useless coding agent work.<p>Lightbulb moment. There\u2019s nothing more noble than destroying the institutions! Let\u2019s take on Claude Code and Codex, we can do it! Piece of cake!<p>And so, Bullet started off as a side project. We used the Claude Code to improve the Claude Code:<p>1. Model routing. Do you regret giving a task to Fable when it could have literally been done by Sonnet?<p>2. Targeted code + context search. We think embedding the whole repo is dumb. We also think sticking the whole context (or compressed context) in chat is dumb. So we do faster and better greps over both.<p>3. Aggressive context hygiene. Tool output is bounded, stale screenshots disappear, we don\u2019t re-read files\u2026the garbage never floods the model.<p>4. Efficient turns. Batch independent investigation, make one surgical edit, then perform one focused verification. Internal measurement showed 16% fewer round trips and 27% lower cost.<p>5. The Flash. We prayed to Barry Allen for speed.<p>And thank the Flash, he gave us speed! On SWE-bench Verified, Bullet resolved 479&#x2F;500 (95.8%) in one attempt, averaging 119s per task, 35\u201367% faster than mini-SWE-agent + Fable&#x2F;Sol depending on task. Full results and methodology here (<a href=\"https:&#x2F;&#x2F;www.codewithbullet.com&#x2F;blog&#x2F;benchmark-results.html\" rel=\"nofollow\">https:&#x2F;&#x2F;www.codewithbullet.com&#x2F;blog&#x2F;benchmark-results.html</a>)<p>Eventually we started using it every day and never went back.<p>Listed above were just some of the things about Claude Code that frustrated us the most, but we are constantly optimizing every day (look at that, maybe we did learn something from our jobs).<p>In our development, the biggest insight was that model speed matters less than reducing round trips. Independent searches, reads, and commands should happen in parallel, while dependent editing and verification stay sequential. One surprising obstacle was code search, small issues like regex-dialect mismatches caused silent misses and sent agents down completely wrong paths, so we built targeted search with fallbacks and bounded context. The most interesting use case so far has been long iterative work (like benchmarks, data pipelines, and evaluation loops), where each step depends on the last and running multiple agents can\u2019t help as much.<p>Here\u2019s the video demo (<a href=\"https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=rWVmG5fRKgE\" rel=\"nofollow\">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=rWVmG5fRKgE</a>)<p>We hope that you guys try out Bullet if you are suffering with speed as much as we were, and we hope it brings you joy, rainbows, and faster responses. And if it\u2019s terrible, let us know it\u2019s terrible (we\u2019re masochists btw)! We&#x27;ll be in the comments all day, you can also contact us at bullet@davidhf.com.<p>You can try it at <a href=\"https:&#x2F;&#x2F;codewithbullet.com\" rel=\"nofollow\">https:&#x2F;&#x2F;codewithbullet.com</a>.<p>P.S: we hid a code on the website, see if you can unlock the secret page at the footer, all built with Bullet", "5": "2026-08-22T18:05:13.313472"}
{"0": 82, "1": "hackernews", "2": "https://www.businessinsider.com/former-google-exec-jeff-dean-valuation-for-new-ai-startup-2026-8", "3": "Jeff Dean has been in talks for a $10B valuation for his new AI startup", "4": "Jeff Dean has been in talks for a $10B valuation for his new AI startup. ", "5": "2026-08-22T18:05:13.321908"}
{"0": 85, "1": "hackernews", "2": "https://ojcp.dev/", "3": "Show HN: OJCP \u2013 an open protocol for agent-consumable job data", "4": "Show HN: OJCP \u2013 an open protocol for agent-consumable job data. Author here!<p>Agents are applying to jobs for people right now, with progressively more volume, and there&#x27;s nothing built for it. So they scrape career pages and fight ATS forms with Playwright&#x2F;Browser Use, which breaks constantly (or they get bot blocked). Employers get buried in applications that don&#x27;t fit, candidates hear nothing back, and the resume is now an AI-written thing that another AI scores (which breaks the existing model entirely, btw).<p>OJCP is MCP tools for search and apply, a manifest at &#x2F;.well-known&#x2F;ojcp.json so agents can find providers, and schemas that extend schema.org instead of replacing it. The playground on the site is a live MCP endpoint, so you can throw calls at it right now.<p>Why a spec at all when models keep getting better at figuring things out? Inference can&#x27;t produce authorization. An agent can work out what a form wants. It can&#x27;t establish that someone consented to this specific submission, and then employer has no way to verify who&#x27;s calling. So TL;DR a more capable agent is also a more capable impersonator.<p>In this model, trust runs both direction. Agents sign requests using the same method that CloudFlare and OpenAI are already using, providers sign their manifests, agents can check against a JWKS, and trust tiers cap how much candidate PII can go to a given provider. Validation happens at consent, so browsing costs nothing and you only pay the verify when the interaction occurs.<p>I&#x27;m the CTO of Recruitics (job advertising) and spent time at LinkedIn before that, so I&#x27;ve been at the intersection of hiring and job search for a while and have felt the pain of both sides.<p>Happy to answer any questions!", "5": "2026-08-22T18:05:13.340286"}
{"0": 86, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49262312", "3": "Ask HN: How do you keep 54 LLM workflows on the right models?", "4": "Ask HN: How do you keep 54 LLM workflows on the right models?. I have a Django app with 54 LLM-backed workflows. Up until recently I&#x27;ve exclusively used Anthropic models via AWS Bedrock but just set up OpenRouter to test the new Gemini models given they seem to match Sonnet&#x2F;Haiku intelligence but with 3-5x output speed. I&#x27;m using Pydantic AI for validation&#x2F;normalization.<p>I currently maintain a registry that describes a workflow&#x27;s purpose, what we&#x27;re optimizing for (intelligence, speed, cost), its eval, and a human-readable bar that must be achieved.<p>I&#x27;m curious what strategies&#x2F;systems people are using to keep track of everything and ensure an optimal model is being used for a given workflow.", "5": "2026-08-22T18:05:13.347629"}
{"0": 87, "1": "hackernews", "2": "https://github.com/katakate/k7d", "3": "Show HN: K7d \u2013 Fork live Kubernetes clusters in <1s \u2013> GRPO-train AI on infra", "4": "Show HN: K7d \u2013 Fork live Kubernetes clusters in <1s \u2013> GRPO-train AI on infra. Hey HN, Gary here.<p>Today I want to present k7d which is an Apache 2.0, tight Rust VMM + shim enabling something not possible before: fast forking of live running virtualized multi-node k8s clusters with surviving of in-flight connections.<p>A 3-VM nodes K8s cluster gets forked in 105ms, and a 50x fork of a 3-VM cluster in 4.1s on a 64GB RAM box.<p>I have two goals here:<p>1) enable large scale GRPO&#x2F;RL training of AI on Kubernetes infra, which IMO is a great playground for reasoning training, besides training a capability that&#x27;s actually useful. And this requires not only fast episode reset (as you need tens of thousand of multi-turn runs during RL post-training) but also greatly benefits from fast forking so you can do parallel branch exploration, rollback, pruning during RL training. Faithful forks also give you byte-identical starts for the G of GRPO, which gives variance reduction across the group.<p>2) enable &lt;3s VM snapshot pause&#x2F;resume&#x2F;fork of sandboxes with Docker-in-VM, for my other project K7 which provides self-hosted infra for VM sandboxes at scale, with a user-friendly CLI &#x2F; API &#x2F; Python SDK, and Kubernetes native.<p>Besides that, k7d is equipped with:<p>- A Tree-shaped API for resource management: as you can guess when you fork, even with optimized CoW-page-sharing, you want to properly manage your resources (memory + disk) and hence you need to know how to evict while things run. So I have tree-shaped logic to keep track of how children share pages with parents, and let your AI agent protect a promising tree branch, evict an unpromising one, or let LRU-ish logic auto-evict when resources get tight. This tree-based logic applies both to single VM sandboxes, and to multi-VM clusters on their own Linux bridge.<p>- Formal verification: of course not all of it, but selected critical subparts of k7d are formally verified: I use Kani for memory arithmetics in the unsafe paths, and Aeneas (with Lean backend) to formally prove the tree-based logic explained above so that eviction never frees a page referenced by a live descendent.<p>- Latencies as CI: I rigorously keep track of latency for most important operations which remain checked&#x2F;enforced via a suite of integration tests.<p>I really tried hard not building my own VMM and first ended up building another backend for K7 than my initial &quot;kfd&quot; (Kata + Firecracker + Devmapper-snapshotter over LVM thin-pool), which I called &quot;kql&quot; for Kata + Qemu + Longhorn. If you know this stack you&#x27;ll guess it right away: Longhorn is great for cross-node replication so I used it to have my snapshots replicated across nodes, so &quot;snapshot resume&quot; always works &#x2F; HA. Qemu here is because Longhorn&#x27;s block storage requirements was incompatible with Firecracker who wants Devmapper-snapshotter, a backend for which I would not want to build myself the cross-node replication logic.<p>But this &quot;kql&quot; backend yielded forks in 45s due to how Longhorn is built, which was too slow for the users who asked me to enable fast forking for K7.<p>So this is what pushed me towards k7d, named as &quot;k7&#x27;s daemon&quot;, its own native VMM and shim, replacing both Firecracker&#x2F;Qemu and Kata at once.<p>This yields VM sandboxes in K7 which you can fork in under 2-3s, and most of this latency is kubelet overhead, as at the VMM level the warm-fork is actually 5ms.<p>One security trade-off: the daemon has to be shared across branches of a same tree: that&#x27;s by design. So you lose Firecracker&#x27;s Jailer per VM. But I could re-build a similar Jailer per tree, which would be sufficient when a tree isn&#x27;t shared across tenants, such as when you use branching for RL training. That&#x27;s just optimizing for something different than what Firecracker does.<p>The codebase is intentionally tight enough to be audited (&lt;30k LOC for VMM + shim) and I linked a deep-dive blogpost series at the top of the README.<p>I hope you guys will enjoy it and I&#x27;d love contributors and critics.<p>Thx!", "5": "2026-08-22T18:05:14.728382"}
{"0": 88, "1": "hackernews", "2": "https://openobserve.ai/blog/openobserve-vs-prometheus-mimir-metrics-benchmark/", "3": "We benchmarked Prometheus, Mimir, and OpenObserve on 1.09M metrics series", "4": "We benchmarked Prometheus, Mimir, and OpenObserve on 1.09M metrics series. ", "5": "2026-08-22T18:05:14.737765"}
{"0": 89, "1": "hackernews", "2": "https://groq.com/newsroom/groq-closes-usd350-million-series-a-building-the-world-s-leading-ai-inference-cloud", "3": "Groq Closes $350M Series A", "4": "Groq Closes $350M Series A. ", "5": "2026-08-22T18:05:14.743162"}
{"0": 91, "1": "hackernews", "2": "https://github.com/JulienMaille/qtscript-qt6", "3": "Show HN: I used an expiring Codex reset to port QtScript to Qt6", "4": "Show HN: I used an expiring Codex reset to port QtScript to Qt6. Hey HN, here is a short story that might be worth sharing.<p>I have an application that has been scriptable with QtScript for years. I can automate and extend it through js, with access to QObjects, signals, slots, properties, etc.<p>Unfortunately, QtScript disappeared with Qt6, so (after a long research) my plan was to migrate to PythonQt. That meant rewriting existing scripts, deploying a Python runtime, and praying that all our scripts would still be feasible with PythonQt.\nI had no particular desire to do all that, so I had been postponing it. The thinkg about QtScript is that, contrary to Qt, online resources are very scarce, you barely find anything but old forum posts from the 2010s<p>Since I&#x27;m on the $20 Codex plan and had 2 resets that was going to expire unused. I thought I might as well spend it on something slightly unreasonable: take the Qt5.15 QtScript sources and see how far I could get on Qt6<p>-&gt; After roughly one hour, I had a smoke test compiling, linking against Qt6 and evaluating js! (Sol&#x2F;Medium did the plan and the coding)<p>Obviously that was not enough to call it a port, so the real work became validation.\nI kept everything as a small patch series, reviewed the changes, rejected some questionable decisions, and treated CI as the authority, something I&#x27;m not realled used to do.\nThe current matrix covers Qt 6.8 LTS and Qt 6.11 on GCC and MSVC, with 29,890 ECMAScript tests and 137 V8 tests passing<p>I (we?) also ported QtScriptGenerator which was important for my use case.\nWhile QtScript is the JavaScript engine, QtScriptGenerator generates the bindings that expose Qt classes and my application classes to JavaScript.\nThis means a script can take an existing button in the application, reconnect it to custom JavaScript logic, create a new QLabel from JavaScript, inject it into an existing QWidget, and send the result there. No rebuild and no new C++ plugin.\nThis is the part I really did not want to lose.<p>-&gt; took the second reset an another hour to have a project that compiles and pass the smoke tests (this time I used Luna&#x2F;xhigh)<p>To me, the interesting AI part was that this kind of old-code archaeology worked surprisingly well with an agent when the work was constrained and split into reviewable steps.\nMy workflow was: plan the change, let it implement one part, review it, test it, reject the strange ideas, continue.<p>So if someone else is still stuck on Qt 5 mainly because of QtScript or QtScriptGenerator I hope this might be useful.<p><a href=\"https:&#x2F;&#x2F;github.com&#x2F;JulienMaille&#x2F;qtscript-qt6\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;JulienMaille&#x2F;qtscript-qt6</a><p><a href=\"https:&#x2F;&#x2F;github.com&#x2F;JulienMaille&#x2F;qtscriptgenerator-qt6\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;JulienMaille&#x2F;qtscriptgenerator-qt6</a><p>cheers!", "5": "2026-08-22T18:05:14.753370"}
{"0": 92, "1": "hackernews", "2": "https://username.md/", "3": "Show HN: Username.md \u2013 a signed, agent-readable identity page you own", "4": "Show HN: Username.md \u2013 a signed, agent-readable identity page you own. Hi HN. I&#x27;m Chris Bergeron, an SRE by profession but a general technologist in my spare time. Projects I&#x27;ve built have been featured in books, frontpaged on news aggregators and chronicled on my blog.<p>Today, I&#x27;m announcing <a href=\"https:&#x2F;&#x2F;username.md\" rel=\"nofollow\">https:&#x2F;&#x2F;username.md</a> !<p>username.md is a single identity URL:  username.md&#x2F;&lt;your name&gt; - that content-negotiates on the `Accept` header.  It serves an HTML profile page; an agent or curl gets Markdown, JSON, JSON-LD, or a JWT.  The same URL serves different representations based on request header. So the page a human reads and the data a machine parses never drift apart.<p>The interesting part is that every response can be signed and verified. Each handle gets an ED25519 keypair (private key AES-GCM-wrapped at rest), publishes a did:web document, and serves RFC 9421 signed responses. Claims about you \u2014 GitHub, a domain you control, an ATProto handle \u2014 are W3C Verifiable Credentials, verified before they&#x27;re shown (e.g. domain ownership via a DNS-TXT challenge). So &quot;this is really them&quot; can be confirmed by a machine instead of taken on faith.<p>It can also bridge the identities you might already use, like ATProto because it supports did:web.  Every username.md handle is a valid Bluesky identity root, and you can even import from Keybase.<p>Why we built it: usernames on every platform are rented. The platform owns the namespace, the verification badge, and the audience.  As agents start conducting business on our behalf, &quot;who is this, and can I prove it&quot; becomes an accountability and governance issue. I wanted an identity surface that&#x27;s mine, portable, and cryptographically verifiable; which anyone can have if they own a domain (mine is <a href=\"https:&#x2F;&#x2F;chrisbergeron.com\" rel=\"nofollow\">https:&#x2F;&#x2F;chrisbergeron.com</a>, feel free to review the metadata).  But I wanted to bring DID:web functionality to the mass market.  You can get a username for yourself, for an agent, or multiple agents.<p>Free tier gets you a public, signed profile at your handle. Pro ($49&#x2F;yr; coming soon!) adds you@username.md email forwarding with per-service aliases, custom CSS, and an MCP endpoint so an agent can query your profile as tools.<p>The stack is a Python identity kernel, per-user keys, CloudFlare in front of the origin.  Infra is hosted at AWS, built multi-region for HA but only one region is live right now.  I&#x27;m a platform and security guy, so the frontend and marketing copy was augmented with AI.<p>I&#x27;m sure you&#x27;ll find bugs and typos but hopefully few inconsistencies.  It&#x27;s an MVP and I would never be able to ship if I kept engineering, tweaking and optimizing. So, here it is for you to enjoy, explore, and build upon - today.<p>I have some really exciting things on the roadmap so I hope you join me and start building on top of username. The capabilities are super exciting and we&#x27;re just getting started.<p>This may be the beginning of a new category:  SSO as a Service.", "5": "2026-08-22T18:05:14.761992"}
{"0": 93, "1": "hackernews", "2": "https://www.certpost.ai/blog/certbot-renewed-nginx-still-serves-old-cert", "3": "Nginx reloaded nothing. Certbot still exited 0", "4": "Nginx reloaded nothing. Certbot still exited 0. ", "5": "2026-08-22T18:05:14.767946"}
{"0": 94, "1": "hackernews", "2": "https://cactuscompute.com/needle", "3": "Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots", "4": "Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots. Hey HN,<p>Henry from Cactus here!<p>We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2.<p>The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens&#x2F;sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens&#x2F;sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges 300-700 on sub-$200 phones such as the Samsung A-Series.<p>On the tool call and mobile device use benchmarks, Needle 2 trades wins with closest small models like LFM2.5 230M and Apple Foundation Model, at 5x to 70x smaller, both at f16 vs Needle 2 at 2bit. Needle is based on Simple Attention Networks from our paper (<a href=\"https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2607.18363\" rel=\"nofollow\">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2607.18363</a>).<p>Edge AI has lately meant Macs and PCs, but that is just 1.5 billion of over 21 billion connected IoT devices in the world today, and in emerging markets most phones ship under $200, no NPU, cheap GPUs. These include budget phones, Raspberry Pis, microcontrollers, wearables, small robots like Reachy Mini, and connected home devices.<p>A conventional transformer of Needle&#x27;s width and depth spends 164 MFLOPs per token, and even one squeezed down to Needle&#x27;s parameter count spends 87, Needle spends 70. Even on a high-end phone, an always-on assistant lives inside a power budget; every MFLOP is milliwatt-hours, and Needle spends 7x to 85x fewer of them per token than the smallest performant LLMs. More about the architecture in the link.<p>When we structure intelligence for consumer devices as functions with typed parameters, the only hard part is mapping a messy sentence onto them; which function, with which values. Our research found that when framed that way, the problem needs no world knowledge and no open-ended prose, which is why 45M parameters suffice.<p>Needle 2 expands to structured extraction where the schema can be passed in-place of tools and the model returns structured output. You can use Needle as a text-classification model with an enum field, as a summarization model by providing a schema that extracts key fields, everything but free-range decode.<p>Every product has its own tool vocabulary and fine-tuning needle helps it achieve frontier-level performance on custom tasks, so using the python package (<a href=\"https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle</a>), Needle can be fine-tuned Needle on a Mac&#x2F;PC in minutes to a few hours, with automated data-generation pipeline, just pass a couple samples.<p>Nonetheless, every response carries a learned confidence score based our Cactus Hybrid technique. If above your threshold, act, below it, escalate to the cloud or bigger model. Combining Needle 2 with a private DeepSeek-v4-Flash deployment works particularly well for enterprise-level tasks at barely any cost, we can help with this setup.<p>We have put a lot of thoughts into Needle 2 but might still be missing quite a lot, please use the playground in the provided link to test Needle and share your thoughts, always appreciated!", "5": "2026-08-22T18:05:14.775055"}
{"0": 95, "1": "hackernews", "2": "https://nicholas-a-hall.github.io/openlanelink/", "3": "Show HN: ESP32 Bowling System technical write-up", "4": "Show HN: ESP32 Bowling System technical write-up. Hi everyone! I&#x27;m the ESP32 Bowling guy from a few weeks ago, back with a technical write-up and a repo link.<p>Since my last post, I&#x27;ve received -tons- of support and encouragement for the project. I&#x27;m absolutely blown away. I didn&#x27;t think there would be much interest!<p>This last week, I spent a ton of time working on the prototype. It can trigger my pinsetter machines and read ball state, there&#x27;s a basic state machine for the game and a React UI for the bowlers. I&#x27;ve also built the prototype uart bridge which connects an ESP32 gateway node to a raspberry pi lane compute module, which acts as a bridge between the ESPNOW mesh and lane-external services.<p>The last major blocker is object detection on an ESP32-CAM module. I could go with a webcam wired to the Pi, but that feels like cheating. I really want the Pi running the state machine, websocket and REST API endpoints only. One of my key constraints for this project is that each node serves one purpose only, and it performs that purpose well. I know an ESP is resource-constrained though, so I&#x27;m not completely sure I can do pin detection onboard the ESP32-CAM. I&#x27;ve thought about ditching vision-based pin tracking entirely, and going with a mmwave radar module instead - then I&#x27;d get ball detection, speed &amp; trajectory for free on top of the pin tracking I actually need to count score. That tech is still somewhat expensive though (especially multiplied by 8 lanes), so for now I&#x27;ll stick with computer vision. Worst case scenario, I&#x27;ll implement the CV piece on the pi with a webcam to unblock myself there... but again, I&#x27;d really like all sensors and actuators to be esp32 nodes.<p>The state machine is interesting - when I started, I didn&#x27;t think about all the bowling game variants out there. There&#x27;s traditional 10-pin, there&#x27;s 9-pin, candlepin, duckpin... OpenLaneLink needed to support all of them. Right now it does basic 10-pin only, but I&#x27;m working on a game configuration piece that takes in JSON definitions for each game type.<p>The UI has been wired to the state machine, and I can drive a simulated game via REST calls to the pi. I also still need the pinsetter node to read machine state from a bank of optocouplers. Once the vision + optocoupler pieces are implemented, the MVP will be complete and I&#x27;ll finally be able to bring my own center&#x27;s Lane 2 back online.<p>In full disclosure: I did use AI assistance for the project, but was very careful not to just vibe-code everything. I needed a research assistant to go fetch relevant information and assist with &quot;hey what&#x27;s the C syntax for such-and-such again?&quot; I review everything an AI suggests, which even led to an interesting &quot;argument&quot; with Claude about how a bowler&#x27;s scoresheet class should be implemented in the game state machine. It insisted individual frame score should be attributes of the overall game, rather than the individual player. Like... what? One day I&#x27;ll post a write-up on how AI helped the project and how it impeded my progress as well.<p>Repo here: <a href=\"https:&#x2F;&#x2F;github.com&#x2F;nicholas-a-hall&#x2F;openlanelink&#x2F;tree&#x2F;breadboard\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;nicholas-a-hall&#x2F;openlanelink&#x2F;tree&#x2F;breadbo...</a>.<p>Feedback and questions welcome, especially on the ESP32-CAM if anyone&#x27;s done onboard CV like this before! The project&#x27;s a bit of a mess right now, and not ready to deploy elsewhere. I&#x27;m definitely looking forward to publishing an alpha release in the next few weeks (after fully battle-testing it in my own facility, of course).", "5": "2026-08-22T18:05:14.783994"}
{"0": 96, "1": "hackernews", "2": "https://slickfast.com/", "3": "Show HN: SlickFast Deterministic Chart/Dash Renderer, No Browser(JSON \u2192 SVG/PNG)", "4": "Show HN: SlickFast Deterministic Chart/Dash Renderer, No Browser(JSON \u2192 SVG/PNG). SlickFast started 6 months ago, and evolved in a super backwards way. I was using lowfruits to look for good SEO keywords, I wanted to make a simple free tool to rank with SEO. I found some great keywords related to graphs&#x2F;charts. I did research for optimizing for SEO. Turns out edge processing &#x2F; using static HTML is super fast and lightweight, great for SEO. I made freepiechartmaker.com. I was really blown away by how fast the site was loading, and how lightweight all the processing was. The site renders changes on the fly, and is much much faster than other sites in the space. I started looking into the tech, pure math rendering, and I saw a lot of openings for what this tech can do. That&#x27;s how SlickFast was made.<p>SlickFast is a JSON input &gt; pure javascript SVG Native render core &gt; with PNG+SVG output. No headless chrome. No library calls. deterministic output. on my local machine(m1 max) it renders 140,000 svg charts a second. PNG @ ~50&#x2F;sec at retina (scale-2) and ~145&#x2F;sec at scale-1, single-core. (we have a benchmark mjs in our github repo)<p>SlickFast can render 47 chart types. Any resolution, aspect ratio, any tile combination, easily deliverable anywhere, and made for agentic workflows. \nOur MCP is free and open source.<p>Tiling allows complex dashboard creation. Each dashboard is only one render. SlickFast API serves URL endpoints. SlickFast is lightweight, we have a hero dashboard demo at slickfast.com&#x2F;deck that is updating every 3 seconds. This is millions of renders, and the cost per month is less than $10.<p>SlickFast was intended to be ready for agentic workflows, and visualizing agentic output as cheap and frictionless as possible.<p>A benefit to the SlickFast deterministic pure math render core, is that edge servers will easily cache images. Millions of people can see a SlickFast endpoint, and only a few renders will actually be used, because the edge servers will happily cache our very small charts.<p>SVG Native chart rendering provides for retina quality, at 50kb image sizes.<p>Our favorite demo is LIVE GitHub dashboards, living right on the repo page. just put a SlickFast URL in your readme. You can see our living dashboard at github.com&#x2F;slickfast&#x2F;slickfast. We have made githubs templates for the public. you can easily add your own dashboard. <a href=\"https:&#x2F;&#x2F;github.com&#x2F;SlickFast&#x2F;github-dashboard-template\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;SlickFast&#x2F;github-dashboard-template</a><p>SlickFast also has telegram deliverable templates. <a href=\"https:&#x2F;&#x2F;github.com&#x2F;SlickFast&#x2F;telegram-weather-template\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;SlickFast&#x2F;telegram-weather-template</a> - if you go to the repo, the weather dash being sent via telegram, is on the repo as a live dashboard - Its updated everyday with a real weather report.<p>Another great demo is our X.com NYC air quality index bot (@NYCAIRREPORT). Slickfast posts a AQI dash on X.com 3 times a day. With the new x API terms, each post is only about 2 cents. The cost on the slickfast side is basically nothing. Our free API tier currently provides 250 renders a month. Enough space for a few bots. Because slickfast was made for agentic use, an agentic was able to design the entire dash and inputs in only a few takes. AI Agents, can nearly one shot most tasks. One or two minor revisions is usually all it takes.<p>Slickfast.com Github.com&#x2F;slickfast&#x2F;slickfast<p>Slickfast is free and open source under AGPL license. I&#x27;m so delighted to provide some real value for free. We have a roadmap with some super cool things coming. Slickfast is very powerful for email, or tying to APIs.<p>Thanks Everyone! I would love to hear your thoughts.", "5": "2026-08-22T18:05:14.790329"}
{"0": 97, "1": "hackernews", "2": "https://www.vincentschmalbach.com/time-serves-ai-bots-a-different-website/", "3": "TIME Is Serving AI Bots a Different Website, with Ads Built In", "4": "TIME Is Serving AI Bots a Different Website, with Ads Built In. ", "5": "2026-08-22T18:05:14.799747"}
{"0": 98, "1": "hackernews", "2": "https://www.wired.com/story/ai-tells-detection-world-series-of-poker-espn/", "3": "AI Detection of Poker Player Bluffing", "4": "AI Detection of Poker Player Bluffing. ", "5": "2026-08-22T18:05:14.809757"}
{"0": 99, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49148320", "3": "What Is Decispher?", "4": "What Is Decispher?. What is Decispher?<p>The decision layer between your team and every AI agent<p>Decispher is the system of record for engineering decisions. It automatically captures the decisions, conventions, constraints, and rationales your team produces every day from Slack, GitHub PRs, and docs, then serves that knowledge to both humans and AI agents the moment they need it.<p>Think of it as a senior engineer who has read every Slack message, every PR, and every architecture discussion your team has ever had, and is always available to answer &quot;why did we do this?&quot; with a cited, accurate answer in under a second.<p>decispher.com", "5": "2026-08-22T18:05:14.813958"}
{"0": 100, "1": "hackernews", "2": "https://github.com/rextio/rextio", "3": "Show HN: Rextio \u2013 Compile Python code natively using Rust with CPython fallbacks", "4": "Show HN: Rextio \u2013 Compile Python code natively using Rust with CPython fallbacks. Hi, HN! Let me introduce my experimental project Rextio.<p>While Python is widely used in fields such as AI and data analysis and offers the advantage of fast coding, it often causes some performance issues because Python is an interpreter language.<p>Of course, there are various alternatives to address Python&#x27;s speed issues. In some cases, Python code is used only for initial prototypes, and the development is re-evaluated in Rust or C&#x2F;C++ upon transitioning to production.<p>However, I&#x27;m experimentally developing this tool called Rextio as a new alternative to solve speed problems without abandoning Python.<p>Rextio improves performance by automatically converting parts of Python hot paths that can be compiled natively (such as type-hinted static code) into Rust code and compiling them into binaries. It then falls back to the rest of the code as is and executes it using CPython. This allows for the complete preservation of the original Python logic while aiming to improve execution speed for specific parts.<p>Roughly speaking, it is as follows:<p>```\ntyped Python project<p>-&gt; analyze supported native candidates<p>-&gt; reject unsafe or unsupported functions<p>-&gt; generate Rust + PyO3 for accepted functions<p>-&gt; generate Python fallback wrappers for the rest<p>-&gt; build import-compatible artifacts\n```<p>Please refer to the documentations in the Git repository for details.<p>Performance improvements are also confirmed when comparing execution with the native binary converted to Rextio versus execution with the original Python code only using CPython.<p>* When running basic Python code converted to native: 57.73\u00d7<p>* When running NumPy converted to native: 2.52\u00d7<p>* When running NetworkX&#x27; Dijkstra algorithm converted to native: 3.68\u00d7<p>* When running `pandas`&#x27; `Series.map` converted to native: 66.14\u00d7<p>* When running PyTorch deep MLP converted to native: 1.02\u00d7 (Speed   improvement limited due to using C extensions in existing Python packages)<p>* When running TensorFlow eager chain converted to native: 1.04\u00d7 (Speed   improvement limited due to using C extensions in existing Python packages)<p>Please refer to the [`rextio-benchmark` Git repository](<a href=\"https:&#x2F;&#x2F;github.com&#x2F;rextio&#x2F;rextio-benchmark\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;rextio&#x2F;rextio-benchmark</a>) for the source code and execution environment information used in the benchmarks that yielded the above results.<p>Rextio has a Core CLI that converts and compiles Python code, and we are developing additional plugins that define rules for converting individual Python packages into Rust code to be attached to it. We have currently developed some plugins for NumPy and others (`rextio-numpy`, `rextio-networkx`, `rextio-pandas`, `rextio-torch`, `rextio-tensorflow`, etc.), and the benchmark results above were obtained using these plugins.<p>Rextio is still an early-stage project with much room for improvement, and compiling Python code into Rust-based native code does not guarantee performance gains in all situations.<p>For instance, if there is frequent data exchange between natively compiled parts and CPython fallbacks, data conversions crossing the boundaries can actually lead to a decrease in speed. However, in such cases, Rextio tries to fall back the entire process to CPython without passing through native code to prevent this degradation.<p>If you have any questions or feedback, I&#x27;ll do my best to answer them and incorporate them into feature improvements.<p>Thank you.", "5": "2026-08-22T18:05:14.820132"}
{"0": 101, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49138466", "3": "Ask HN: How are you using AI to learn?", "4": "Ask HN: How are you using AI to learn?. It&#x27;s time to start learning advanced topics again and I thought it&#x27;d be at least interesting to take advantage of LLM&#x27;s to help me learn better or faster.<p>I&#x27;ve read here that some of you are using &quot;socratic skills&quot; where you give the LLM some content and then you get a series of questions in order to learn but I don&#x27;t think it&#x27;s the only approach to this.<p>I&#x27;ll start tackling the &quot;Attention is all you need&quot; paper and then I want to go deeper.<p>What&#x27;s your approach?", "5": "2026-08-22T18:05:14.829515"}
{"0": 105, "1": "hackernews", "2": "https://www.bloomberg.com/news/articles/2026-08-12/ai-startup-cognition-in-new-funding-talks-at-40-billion-value", "3": "AI Startup Cognition in New Funding Talks at $40B Value", "4": "AI Startup Cognition in New Funding Talks at $40B Value. ", "5": "2026-08-22T18:05:16.063429"}
{"0": 106, "1": "hackernews", "2": "https://www.businessinsider.com/weave-funding-tokenmaxxing-startup-y-combinator-ai-coding-spend-developers-2026-7", "3": "A YC Startup That Wants to Kill Tokenmaxxing Just Raised $13.5M", "4": "A YC Startup That Wants to Kill Tokenmaxxing Just Raised $13.5M. ", "5": "2026-08-22T18:05:16.068638"}
{"0": 107, "1": "hackernews", "2": "https://www.cnbc.com/2026/04/27/deepmind-ineffable-intelligence-record-seed-funding-nvidia-google.html", "3": "Ex-DeepMind David Silver Raises $1.1B for AI Startup Ineffable", "4": "Ex-DeepMind David Silver Raises $1.1B for AI Startup Ineffable. ", "5": "2026-08-22T18:05:16.073789"}
{"0": 108, "1": "hackernews", "2": "https://restofworld.org/2026/edtech-funding-collapse-k12-startups-ai-workforce/", "3": "The global edtech boom is fading as investors look elsewhere", "4": "The global edtech boom is fading as investors look elsewhere. ", "5": "2026-08-22T18:05:16.079578"}
{"0": 110, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=47718079", "3": "Ask HN: Will we ever do something about patent trolls?", "4": "Ask HN: Will we ever do something about patent trolls?. Some of you may have seen a project like Mycroft AI (nothing to do with LLMs!) from back in around 2017 or so, it was an open source Alexa alternative, even had skills and everything. I kept wondering why it quietly disappeared, and the repo was archived. Turns out a patent troll hit them hard, drained them of all their funds in court, till they shut down.<p>https:&#x2F;&#x2F;patentprogress.org&#x2F;2024&#x2F;05&#x2F;another-startup-bites-the-dust-courtesy-of-patent-trolls&#x2F;<p>https:&#x2F;&#x2F;www.theregister.com&#x2F;2023&#x2F;02&#x2F;13&#x2F;linux_ai_assistant_killed_off&#x2F;<p>Then there&#x27;s many others, and I&#x27;m sure I&#x27;m missing some, but here&#x27;s a list of companies &#x2F; projects hit by patent trolls with various outcomes:<p>That time some company shut down a free &#x2F; open source site with a patent that came 10 years after the technology already existed.<p>https:&#x2F;&#x2F;www.techdirt.com&#x2F;2014&#x2F;11&#x2F;19&#x2F;patent-troll-kills-open-source-project-speeding-up-computation-erasure-codes&#x2F;<p>https:&#x2F;&#x2F;x.com&#x2F;JamesBessen&#x2F;status&#x2F;532906754364149760<p>Here&#x27;s one where the courts did the right thing for once, against the Gnome project (note Gnome probably has better funding than most of these other smaller projects).<p>https:&#x2F;&#x2F;opensource.org&#x2F;blog&#x2F;gnome-patent-troll-stripped-of-patent-rights<p>Here&#x27;s one I didn&#x27;t even know about, if you ever posted a job ad on LinkedIn a company might have contacted you threatening to sue you for violating their patent (Whiskey Tango F....):<p>https:&#x2F;&#x2F;www.zdnet.com&#x2F;article&#x2F;open-source-fights-back-we-wont-get-patent-trolled-again&#x2F;<p>A few years ago, CloudFlare did have their own brawl with some as well, and it looks like thankfully they did not backdown:<p>https:&#x2F;&#x2F;blog.cloudflare.com&#x2F;three-new-winners-of-project-jengo-and-more-defeats-for-the-patent-troll&#x2F;<p>Small mobile app developers were hit for having links to payment providers:<p>https:&#x2F;&#x2F;www.bbc.com&#x2F;news&#x2F;technology-14682700<p>Apple was sued over Facetime, of all things, and lost:<p>https:&#x2F;&#x2F;www.bbc.com&#x2F;news&#x2F;technology-20236114<p>I ask here, because this is one of the most pro-startup pro-hacker anti-patent troll communities on the web, does anyone know if there&#x27;s any org doing something about patent trolling? It&#x27;s ridiculous that people who provide no value to society can just waste court time and siphon funds from hard working people just because they bought a patent. The patent framework needs to be reworked to force a patent holder to prove they are actually using their patent to build something, or show they have built something that is on the market, and that the violations endanger their efforts, if they cannot produce this, they should be made to pay all court and lawyer fees.<p>Just my thoughts, I&#x27;m sure there&#x27;s better ways to handle it, but if the rules are changed to stop patent trolls from basically extorting hard working people who actually invent and produce things, I think we could see a lot more. In the giant shift of AI since Mycroft was birthed, I can&#x27;t imagine how much more advanced Mycroft would have been by now.<p>I hate when simple things stifle innovation. I&#x27;m an innovation junkie, I want to see the future we all saw as kids in cartoons and scifi that fascinated us, but we&#x27;re often held back by bad actors.<p>Are there orgs or initiatives? Is anyone ... and I hate the word, but it really should be done, is anyone lobbying to protect inventors from patent trolls? Because they do more harm than good, and they benefit no one but themselves, while again, stifling innovation.", "5": "2026-08-22T18:05:16.093622"}
{"0": 111, "1": "hackernews", "2": "https://www.nytimes.com/2026/03/12/technology/axiom-ai-code-funding.html", "3": "A.I. Writes Buggy Code. A Silicon Valley Startup Wants to Fix It", "4": "A.I. Writes Buggy Code. A Silicon Valley Startup Wants to Fix It. ", "5": "2026-08-22T18:05:16.104865"}
{"0": 112, "1": "hackernews", "2": "https://www.bloomberg.com/news/articles/2026-03-10/yann-lecun-s-new-ai-startup-raises-1-billion-in-seed-funding", "3": "Yann LeCun's AI startup raises $1B seed round", "4": "Yann LeCun's AI startup raises $1B seed round. ", "5": "2026-08-22T18:05:16.111845"}
{"0": 113, "1": "hackernews", "2": "https://www.bloomberg.com/news/articles/2026-02-17/china-ai-startup-moonshot-seeks-10-billion-value-in-new-funding", "3": "China AI Startup Moonshot Seeks $10B Value in New Funding", "4": "China AI Startup Moonshot Seeks $10B Value in New Funding. ", "5": "2026-08-22T18:05:16.117945"}
{"0": 114, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=46967486", "3": "Ask HN: What makes early-stage AI accelerators useful (and what doesn't)?", "4": "Ask HN: What makes early-stage AI accelerators useful (and what doesn't)?. Hi HN \u2014 we recently launched the Berkeley Xcelerator (https:&#x2F;&#x2F;rdi.berkeley.edu&#x2F;xcelerator), a non-dilutive accelerator program run by Berkeley RDI (https:&#x2F;&#x2F;rdi.berkeley.edu&#x2F;) for pre-seed and seed-stage teams building in AI and agentic AI. We\u2019d love to get some feedback from the community!<p>Over the past three years, Berkeley Xcelerator has supported 110+ teams across AI, cybersecurity, and decentralized technologies, whose founders have gone on to raise $650M+ in follow-on funding, spanning 100+ countries.<p>Some concrete details about the Xcelerator itself:<p>- The program is non-dilutive (no equity taken)<p>- Open to pre-seed and seed-stage AI &#x2F; agentic AI startups<p>- No UC Berkeley affiliation required<p>- Selected teams receive support through Berkeley RDI\u2019s research community and ecosystem partners<p>- Enablement includes cloud, GPU, and API credits from industry partners (including Google Cloud, Google DeepMind, OpenAI, and Nebius, with more to be announced)<p>- The program culminates in a Demo Day at the Agentic AI Summit (Aug 1\u20132, 2026) at UC Berkeley, where we are expecting 5,000+ in-person attendees<p>Here\u2019s what we\u2019d really like input on:<p>- If you\u2019ve built or joined an early AI startup, what actually helped you most early on?<p>- If you\u2019ve done an accelerator, what helped and what was a waste of time?<p>- For technically deep projects (infra, agentic systems, safety-sensitive work), what kinds of feedback or structure mattered most before product-market fit?<p>If you\u2019d like to apply to the Berkeley Xcelerator, applications are open through the end of February. (https:&#x2F;&#x2F;forms.gle&#x2F;KjHiLAHstAvfHdBf7)", "5": "2026-08-22T18:05:16.125803"}
{"0": 115, "1": "hackernews", "2": "https://rdi.berkeley.edu/xcelerator", "3": "Show HN: Berkeley Xcelerator \u2013 early-stage AI and agentic AI accelerator", "4": "Show HN: Berkeley Xcelerator \u2013 early-stage AI and agentic AI accelerator. Hi HN \u2014 we recently launched the Berkeley Xcelerator (<a href=\"https:&#x2F;&#x2F;rdi.berkeley.edu&#x2F;xcelerator\" rel=\"nofollow\">https:&#x2F;&#x2F;rdi.berkeley.edu&#x2F;xcelerator</a>), a non-dilutive accelerator program run by Berkeley RDI (<a href=\"https:&#x2F;&#x2F;rdi.berkeley.edu&#x2F;\" rel=\"nofollow\">https:&#x2F;&#x2F;rdi.berkeley.edu&#x2F;</a>) for pre-seed and seed-stage teams building in AI and agentic AI.<p>Over the past three years, Berkeley Xcelerator has supported 110+ teams across AI, cybersecurity, and decentralized technologies, whose founders have gone on to raise $650M+ in follow-on funding, spanning 100+ countries.<p>Some concrete details about the Xcelerator itself:<p>- The program is non-dilutive (no equity taken)<p>- Open to pre-seed and seed-stage AI &#x2F; agentic AI startups<p>- No UC Berkeley affiliation required<p>- Selected teams receive support through Berkeley RDI\u2019s research community and ecosystem partners<p>- Enablement includes cloud, GPU, and API credits from industry partners (including Google Cloud, Google DeepMind, OpenAI, and Nebius, with more to be announced)<p>- The program culminates in a Demo Day at the Agentic AI Summit (Aug 1\u20132, 2026) at UC Berkeley, where we are expecting 5,000+ in-person attendees<p>Happy to answer questions here! Your feedback and participation are incredibly valuable to us.<p>Applications are open through the end of February (<a href=\"https:&#x2F;&#x2F;forms.gle&#x2F;KjHiLAHstAvfHdBf7\" rel=\"nofollow\">https:&#x2F;&#x2F;forms.gle&#x2F;KjHiLAHstAvfHdBf7</a>)", "5": "2026-08-22T18:05:16.130755"}
{"0": 116, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=46917293", "3": "Ask HN: Best option for hosted agent in 2026?", "4": "Ask HN: Best option for hosted agent in 2026?. Hey everyone. I have a simple question and have had a very tough time finding the answer.<p>Our startup wants to build and use an agent. I want to give the agents skills, access to our APIs, etc. I do <i>not</i> want to reinvent any wheels, I do not want to reinvent an agent loop.<p>Ideally, we&#x27;d effectively have a hosted version of Claude Code with company specific skills, an API access layer, etc. In the way that Claude Code &#x2F; Cursor has access to our codebase + skills, thats effectively what we want and then start exposing it in a limited fashion to our internal team and then later to our customers. And, for example, add specific features to our existing product such as &quot;summarize this with AI&quot; but instead of that api call going to a model provider, it would go to &#x27;our&#x27; agent.<p>Does this make sense?<p>Could someone please point me in the right direction on what the best &#x2F; simplest solution would be? I feel like it should be easy and obvious but I&#x27;m struggling to find a clear path forward. Thank you all! - Mike", "5": "2026-08-22T18:05:16.140078"}
{"0": 117, "1": "techcrunch", "2": "https://techcrunch.com/2026/08/22/harvards-699-startup-bootcamp-offers-ai-avatars-of-its-instructors/", "3": "Harvard\u2019s $699 startup bootcamp offers AI avatars of its instructors", "4": "Harvard\u2019s $699 startup bootcamp offers AI avatars of its instructors. In the HBS Foundry program, AI avatars provide feedback during practice pitches and board meetings.", "5": "2026-08-23T06:45:02.574408"}
{"0": 131, "1": "hackernews", "2": "https://meetless.ai", "3": "Show HN: Active Source of Truth for Your Coding Agents", "4": "Show HN: Active Source of Truth for Your Coding Agents. Howdy! Happy Saturday everyone!<p>As a solo founder, I have always tried to maximize my speed by letting coding agents build as much as possible in parallel. However, as an engineer, I don&#x27;t trust that AI will always make the right decisions and work with the right context. In the past, I always needed to click through my sessions to glance at the AI&#x27;s output, try to understand what it was doing, and hopefully steer it or stop it in time.<p>As a result, the maximum number of concurrent sessions I could manage at once was only 4. I didn&#x27;t want to be the bottleneck, so I built Meetless Agent (MLA). It basically does what I had to do manually before:<p>- Monitors the coding agent&#x27;s tasks and actions to supply it with the correct, up-to-date context.<p>- Continuously reconciles running information (such as provided&#x2F;tagged documentation, the agent&#x27;s output, and the agent&#x27;s decisions) to actively maintain the source of truth at all times.<p>- Keeps track of the current rules for the repo, and if an action triggers a registered rule, the rule is injected into the agent context.<p>My benchmarks show that running coding agents with the help of an active monitor improves quality and accuracy, consumes fewer tokens, and finishes faster: <a href=\"https:&#x2F;&#x2F;research.meetless.ai&#x2F;stale-context&#x2F;\" rel=\"nofollow\">https:&#x2F;&#x2F;research.meetless.ai&#x2F;stale-context&#x2F;</a><p>Of course, the agent alone can&#x27;t decide the source of truth; it requires human review and decisions for contradictions, etc. But for the most part, it can safely build a consistent ontology of the current source of truth.<p>From this, I want to build an AI layer to maintain the source of truth across the business, so I will release more connectors for Slack, Jira, Confluence, etc., soon. The goal is for this AI to assist in every part of the business. Eventually, the same coordination layer will understand that a decision made in Slack affects a Jira task, a document, an email conversation, and what a coding agent should do next without every tool becoming another isolated memory silo.<p>The coding agent connector is open source at:\n<a href=\"https:&#x2F;&#x2F;github.com&#x2F;Meetless&#x2F;mla\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;Meetless&#x2F;mla</a><p>I am looking forward to your feedback!", "5": "2026-08-23T06:45:13.907315"}
{"0": 137, "1": "hackernews", "2": "https://gene-inspector.pro/", "3": "Show HN: Gene Inspector Pro \u2013 I built a tool for exploring your own genome", "4": "Show HN: Gene Inspector Pro \u2013 I built a tool for exploring your own genome. Hi HN, I\u2019m Sergey.<p>When my son was diagnosed with several health conditions, a doctor told us to \u201clearn to live with it.\u201d I refused and started looking for answers.<p>I spent the next seven years teaching myself genetics and cellular biology. That search led me to build Gene Inspector and have my family\u2019s DNA sequenced. I now study genomic data from 14 family members across three generations.<p>Gene Inspector turns raw DNA data into a shortlist of findings people can investigate. Instead of searching through millions of variants, users can quickly find potentially interesting ones, explore how they relate to health topics, pathways, or medications, and see the evidence behind each result.<p>Gene Inspector supports data from common DNA tests, Whole Exome Sequencing, and Whole Genome Sequencing. Findings are annotated using public genetic sources and linked to the underlying evidence.<p>I also built a pipeline that processes open-access papers, finds claims about specific variants, and links each claim to its source. This lets users check the research themselves instead of trusting a black-box interpretation.<p>Because most people are not geneticists, I\u2019m now building an AI research agent Diana that explains what users are seeing and helps connect the dots.<p>Diana will wear two &quot;hats&quot;. Her &quot;geneticist&quot; hat will make her strict and skeptical, focusing on stronger clinical evidence. Her &quot;functional medicine&quot; hat will look more broadly at metabolism, enzymes, pathways, and common variants - just like a functional MD would do.<p>Diana already supports text and real-time voice conversations but is currently available only through an invitation-only beta.<p>I built most of Gene Inspector&#x27;s by hand before the AI wave.<p>I\u2019d appreciate one simple piece of feedback: would you find Gene Inspector useful?", "5": "2026-08-23T06:45:13.938019"}
{"0": 138, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49336154", "3": "Ask HN: Who needs funding for DB research?", "4": "Ask HN: Who needs funding for DB research?. This is the sibling post of \u201cWho wants to fund DB research?\u201c [1], please see that post for context. In short, if you are working on non-AI related DB topics and are looking for funding, leave a comment below with a short summary of your work. I\u2019m focusing on non-AI areas because their funding has been declining largely due to the spending on AI. Format:<p>Name, Affiliation<p>Research topics<p>[1]: https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=49336147", "5": "2026-08-23T06:45:13.943771"}
{"0": 139, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49336147", "3": "Ask HN: Who wants to fund DB research?", "4": "Ask HN: Who wants to fund DB research?. You may have heard there\u2019s a funding crisis for science in general, and for computer science in particular. Within CS, despite billions of dollars being spent for AI research, there is less and less funding for traditional areas. I work in Databases, and have seen many colleagues switching to AI-related work like vector DBs and agent systems, even though there have been lots of exciting breakthroughs on core DB topics like WCOJ and Datalog, to name a few. As I write this, a post on DuckDB v2.0 is top 1 of Hacker News, with many users commenting it\u2019s one of their favorite pieces of software. DuckDB is the direct result of decades of DB research [1].<p>HN has a tradition of \u201cwho\u2019s hiring\u201d posts, and I figured we could also do a \u201cwho\u2019s funding\u201d post. If you are interested in funding DB research, feel free to leave a comment, or reach out to me directly [2]. If you are working in DB and are looking for funding, leave a comment under this post: [3].<p>[1]: https:&#x2F;&#x2F;duckdb.org&#x2F;why_duckdb#standing-on-the-shoulders-of-giants<p>[2]: https:&#x2F;&#x2F;remy.wang<p>[3]: https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=49336154", "5": "2026-08-23T06:45:13.951600"}
{"0": 140, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49329291", "3": "Ask HN: \u20ac300/mo to capture work you're doing. Tips how to reach right people?", "4": "Ask HN: \u20ac300/mo to capture work you're doing. Tips how to reach right people?. Hey Everyone,<p>My name is Salan and I&#x27;m a Researchrer at https:&#x2F;&#x2F;pdoom.org&#x2F; an early-stage AI Research Lab building LLMs for long-horizon tasks in Europe. We were funded by the German Government (SPRIND).<p>Currently, we are trying to acquire data, specifically from people who do open-source work&#x2F;research. Therefore, we pay people \u20ac300 per month to capture their digital workflows.<p>We also built a custom tool that is privacy-preserving specifically for capturing these workflows: Get paid to record your work \u00b7 p(doom)<p>Since we have a high quality bar and need really strong people in their fields, finding participants hasn\u2019t been easy. Do you guys have any tips or ideas on how we can successfully reach these profiles?<p>Any help is highly appreciated!", "5": "2026-08-23T06:45:13.957842"}
{"0": 141, "1": "hackernews", "2": "https://www.seattletimes.com/seattle-news/science/ai-is-finding-sperm-where-doctors-couldnt/", "3": "AI is finding sperm where doctors couldn't", "4": "AI is finding sperm where doctors couldn't. ", "5": "2026-08-23T06:45:13.962376"}
{"0": 142, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49321400", "3": "Ask HN: What tools are you using for human code review of AI-assisted code?", "4": "Ask HN: What tools are you using for human code review of AI-assisted code?. A good proportion of us and our colleagues are now churning out agent-assisted code at an incredible rate, with some of it that is actually good, and a lot that is not so good. I&#x27;m personally finding that the real quality gate for our projects is now how thoroughly the generated code was human reviewed to ensure that it is not just correct, but architecturally sensible.<p>AI code review tools like coderabbit and copilot, or even pointing claude code at a PR are all generally pretty good at finding bugs and style nits, but less good at finding duplicate code, module cross coupling, bad separation of concerns, and so on, even if prompted to do so.<p>I&#x27;m finding that github&#x27;s PR interface is not really cutting it for me, it was janky even when the reviews were small, but now at the size they&#x27;re at, it is becoming unmanageable.  Add to that the extra noise of mixing in agent reviews, and people &quot;meat-proxying&quot; in copy-pasted agent output, and it&#x27;s getting pretty noisy and difficult to navigate.<p>What have you all found that works well for streamlining human review of AI assisted code? Tools and process suggestions are welcome.", "5": "2026-08-23T06:45:13.967881"}
{"0": 143, "1": "hackernews", "2": "https://news.bloombergtax.com/tax-insights-and-commentary/tax-law-is-funding-the-ai-infrastructure-boom-not-creating-it", "3": "Tax Law Is Funding the AI Infrastructure Boom, Not Creating It", "4": "Tax Law Is Funding the AI Infrastructure Boom, Not Creating It. ", "5": "2026-08-23T06:45:13.972578"}
{"0": 144, "1": "hackernews", "2": "https://blog.audn.ai/posts/audn-vs-codex-vs-aikido-juice-shop-comparison", "3": "Different SAST/DAST scanners only share %1.8 of common findings on same source", "4": "Different SAST/DAST scanners only share %1.8 of common findings on same source. ", "5": "2026-08-23T06:45:13.978320"}
{"0": 161, "1": "hackernews", "2": "https://arxiv.org/abs/2608.19202", "3": "Active Inference as Context Acquisition for AI Agents", "4": "Active Inference as Context Acquisition for AI Agents. ", "5": "2026-08-23T06:45:22.075857"}
{"0": 176, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=49405520", "3": "Why can AI generate Super Mario but not a wedge ramp for my robot vacuum?", "4": "Why can AI generate Super Mario but not a wedge ramp for my robot vacuum?. I&#x27;ve been puzzled by something: AI generation can produce an elaborate\n  figurine, a cartoon character, even a convincing Super Mario \u2014 yet it\n  can&#x27;t reliably make a simple wedge ramp so my robot vacuum can climb a\n  step.<p><pre><code>  For context: I bought a Bambu P2S but can&#x27;t model. I tried the &quot;describe\n  it and get a model&quot; AIs \u2014 the output is unusable, you can&#x27;t adjust it,\n  it&#x27;s never quite what I meant. I tried having an agent write Python to\n  build geometry directly \u2014 it tops out at simple primitives.\n \n  What finally worked: geometric decomposition. I break a complex part into\n  ordered, grouped steps, describe each as a small spec, and let an agent\n  execute them in Blender (via blender-mcp). That process turned out to\n  abstract into a small engine \u2014 the key insight being it converts the 3D\n  spatial reasoning LLMs are bad at, into the structured code they&#x27;re good\n  at. I wrote it up here: https:&#x2F;&#x2F;github.com&#x2F;zhuchaokn&#x2F;spec-3d-model\n \n  My questions:\n  - Why is &quot;functional part&quot; generation so much weaker than\n  &quot;figurine&#x2F;aesthetic&quot; generation? Is it data (no parametrized-CAD training\n  sets), representation (mesh vs B-rep), or evaluation (nobody benchmarks\n  &quot;does it print &#x2F; is it watertight&quot;)?\n  - Is &quot;turn 3D modeling into code for an LLM&quot; the right framing, or am I\n  missing something better?</code></pre>", "5": "2026-08-23T06:45:23.818841"}
{"0": 190, "1": "hackernews", "2": "https://neolabs.fyi/", "3": "Show HN: Neolabs.fyi \u2013 100 new AI labs by research area, valuation, and more", "4": "Show HN: Neolabs.fyi \u2013 100 new AI labs by research area, valuation, and more. ", "5": "2026-08-23T06:45:23.886403"}