text stringlengths 0 2.68k |
|---|
• Adam Stacoviak describes the logs as containing the "what and why", with the builder logs capturing the journey of the developer during implementation |
• He also mentions the knowledge base, which contains institutional knowledge and is used to store long-term knowledge |
• The conversation touches on the benefits of using Amp, including increased productivity and the ability to work asynchronously |
• Adam Stacoviak and Beyang Liu discuss the potential of Amp to be used in a pull request workflow, providing an audit trail of the plan that generated the code |
• They also mention the tool's built-in notification sound, which has become Pavlovian for Adam Stacoviak, signaling that it's time to check on the status of a task. |
• Amp is a coding agent used by the team, with around 80-90% of code generated using it |
• Iterations on prompts, tool definitions, and sub-agent combinations have improved Amp's performance |
• Adam Stacoviak created a product manager role using Amp and had it review output from other agents, including Claude Code |
• Beyang Liu explains the pricing model and the trade-off between cost and quality |
• Cheaper pricing may lead to model quality degradation and a perverse incentive to nerf models |
• Time saved and additional value created by using Amp are considered more important than cost |
• Adam Stacoviak's perspective on using coding agents as a skill to be learned and flexed |
• Importance of context and clarity in relationships, including with machines like Amp |
• Discussion of leveraging Amp for specific tasks and workflows |
• The importance of efficient token usage with Amp |
• The difference in behavior patterns between senior engineers and non-technical users |
• Recommendations for efficient token usage, including creating targeted and short threads |
• The use of Amp to improve a Bash script for archiving and compressing media files |
• The development of a new tool, 7zarch, for advanced archiving and compression |
• The user is concerned about using Amp efficiently and is worried about the context window collapsing and forgetting context. |
• Beyang Liu explains that the user's approach is not uncommon and that the company is still learning about how to use Amp effectively. |
• Beyang Liu mentions that the context window limit was previously around 200k, but now it's much higher and the quality has improved beyond 70k. |
• He suggests that the user should start fresh for each task and not accumulate too much context, as it can lead to confusion and degradation in performance. |
• The user agrees that they were using Amp inefficiently and that the company's approach is more intuitive and better suited to professional engineers. |
• Beyang Liu mentions the tension between being prescriptive and allowing users to use Amp in their own way, and suggests that the company will provide more visual indicators and best practices in the future. |
• The user suggests that the company should not change the way Amp works, but rather add a slash command for users who want to read documentation and use an alternate version. |
• The context window and how to leverage threads is a black box for some developers |
• The context window is analogous to the human brain's working memory and has limitations |
• Overloading the context window can lead to latency, degraded quality, and confusion |
• MCP servers can inject irrelevant tool definitions, adding to the context window and causing issues |
• Using roles and props can be more efficient and reduce token costs |
• Threads can be composed like functions, allowing for more efficient agent flow |
• Analogies between human brain function and agent flow can help developers understand the context window and threads |
• The conversation turns to discussing the inception of Amp and the process of raising an agent. |
• Design constraints of AI models in the application |
• Building a spike to experiment with new technology |
• Discovering new workflows and capabilities of AI agents |
• Importance of thinking from first principles and relearning assumptions |
• Sharing learnings and insights with the user community through the podcast |
• Discussing the podcast's format and reach |
• Advice on publishing and promoting content on YouTube |
• Encouraging the hosts to maintain the fun and whimsical tone of their podcast |
• Praising the work of Thorsten, a Sourcegraph employee, on the podcast and as a writer |
• Discussing the growth and progress of Sourcegraph and its mission to help developers |
• Describing the hosts' passion for building developer tools and their desire to continually improve and innovate |
• Sharing personal anecdotes and experiences with the hosts and Sourcegraph's journey |
• Discussing the intersection of technology and human experience, and the joy of coding and software development |
• The contrast between the beauty of nature and the monotony of coding |
• The development of AI and its potential to capture human-like intelligence and reasoning |
• The concept of a "universal pattern matcher" and its ability to fit any observable pattern |
• The hype and skepticism surrounding AI, with some viewing it as a solution to all problems or a threat to human existence |
• The potential for AI to be a useful tool, rather than a replacement for humans |
• The need to approach AI with a nuanced perspective, recognizing its limitations and capabilities. |
• Pattern recognition and automation in technology |
• Mindset for approaching new technologies: exploration and curiosity vs. skepticism and criticism |
• Benefits of using coding agents for building tools and applications outside one's expertise |
• Open source software and its potential future with the rise of coding agents |
• Impact of coding agents on the use and development of libraries and APIs |
• Discussion of using Amp's agentic coding tools without needing to choose a specific model |
• Importance of sampling multiple coding agents to find the best fit |
• Future plans for experimenting with new models and reducing latency in Amp |
• Upcoming release of Amp on September 17th |
• Discussion of making the Raising Agents podcast more frequent and production-level |
**Adam Stacoviak:** Beyang Liu, welcome back to the Changelog. Let's go as deep as humanly possible on Amp. Not Sourcegraph necessarily, but Amp. What do you think? |
**Beyang Liu:** Cool. Yeah, it sounds good to me, and thanks for having me back on the show, Adam. |
**Adam Stacoviak:** What is Amp? |
**Beyang Liu:** Yeah, I don't think it takes too much explanation. Amp is a coding agent. So most people - I would imagine your audience - know what that is at this point. But if you've been living under a rock for the past six months, a coding agent is essentially an AI-powered program that takes natural language inst... |
And then Amp, in particular, among the landscape of coding agents, is distinguished by the fact that -- I think we're the only coding agent that I've come across that has this approach of... So we're a multi-model, we use multiple LLMs, which is not distinctive. There's a lot of coding agents that use a variety of unde... |
So there's no toggling on different specific models. It's more "Oh, for this particular feature or sub-capability of Amp we're going to use this model, because it has the right characteristics in terms of latency, intelligence, and competency around one particular task." |
**Adam Stacoviak:** Maybe just a side tangent, which I didn't really plan to do until this very moment... But I was thinking back to the last time we had a conversation, and I think at the very end -- I'm not even sure if it ended up on the air or not. So this may have just been a side conversation between you and I, a... |
**Beyang Liu:** \[laughs\] |
**Adam Stacoviak:** So obviously we're talking about Amp... I don't know where Cody went. I've literally forgotten about Cody, until this very moment... And that feedback I gave to you, which was "Hey, I'd love to play with your stuff, but I don't feel I know how to do it." I think I have a Sourcegraph account, I've tr... |
**Beyang Liu:** Oh, you've run into auth issues with Amp? |
**Adam Stacoviak:** No, no, no, this was Cody, back when we talked -- I don't even know when we talked. Maybe 8 months ago, 10 months ago, or something that. |
**Beyang Liu:** Yeah. |
**Adam Stacoviak:** So where's Cody? Did Cody just go away? |
**Beyang Liu:** Cody is still alive and well in the enterprise. |
**Adam Stacoviak:** Okay... |
**Beyang Liu:** But we've sort of moved away from it for non-enterprise use cases. So there's still plenty of enterprise customers that use it quite heavily for, I would say non-agentic AI coding assistance... And that's still a common feature among a lot of large companies. There's some companies out there that even h... |
**Adam Stacoviak:** Oh, gosh... |
**Beyang Liu:** You can't say agent inside some orgs. |
**Adam Stacoviak:** What's the alternative word to agent if you can't say agent? |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.