harsh-jos/gemma-3-1b-it-linkedin-natural
Text Generation • Updated • 8
Datasets:
prompt stringlengths 43 95 | response stringlengths 141 1.65k | text stringlengths 259 1.81k |
|---|---|---|
Write a LinkedIn post about: An engineering rulebook I built for AI coding agents and humans | An engineering rulebook I built for AI coding agents and humans (We're now flexing md files 😄)
It covers the usual fundamentals. Nothing groundbreaking for a software engineer.
But I built it anyway.
I'm working on a big project where we do 100% agentic coding with Claude Code. The codebase works. But maintenance is ... | <start_of_turn>user
Write a LinkedIn post about: An engineering rulebook I built for AI coding agents and humans<end_of_turn>
<start_of_turn>model
An engineering rulebook I built for AI coding agents and humans (We're now flexing md files 😄)
It covers the usual fundamentals. Nothing groundbreaking for a software engin... |
Write a LinkedIn post about: Why I stopped using jargon | I stopped using jargon. My posts got better.
"Leveraging synergistic AI paradigms" became "using Claude to write code".
Same idea, but now my friend who is a designer gets it.
Jargon hides vague thinking. Simple words force you to be precise.
If a junior wouldn't get it, rewrite it. Clarity is kindness. | <start_of_turn>user
Write a LinkedIn post about: Why I stopped using jargon<end_of_turn>
<start_of_turn>model
I stopped using jargon. My posts got better.
"Leveraging synergistic AI paradigms" became "using Claude to write code".
Same idea, but now my friend who is a designer gets it.
Jargon hides vague thinking. Si... |
Write a LinkedIn post about: The infinite shelf is a trap | The infinite shelf is a trap.
AI lets you build anything, so you build everything. Then you maintain nothing.
Constraints were a gift. Keep some voluntarily. | <start_of_turn>user
Write a LinkedIn post about: The infinite shelf is a trap<end_of_turn>
<start_of_turn>model
The infinite shelf is a trap.
AI lets you build anything, so you build everything. Then you maintain nothing.
Constraints were a gift. Keep some voluntarily.<end_of_turn>
|
Write a LinkedIn post about: Early access lessons from Kiro | Got Kiro early. Thought I'd be more productive. First week I was slower.
I was prompting like I was talking to a human. Too much context, too polite.
Kiro wants precision, not pleasantries. Tell it what to do, what to read, what done looks like.
Once I adapted, speed doubled.
Early access isn't about being first. I... | <start_of_turn>user
Write a LinkedIn post about: Early access lessons from Kiro<end_of_turn>
<start_of_turn>model
Got Kiro early. Thought I'd be more productive. First week I was slower.
I was prompting like I was talking to a human. Too much context, too polite.
Kiro wants precision, not pleasantries. Tell it what t... |
Write a LinkedIn post about: Why intuition for what AI grasps matters | My intuition for what Claude grasps instantly vs what needs more context changed everything.
It grasps boilerplate, patterns, and transforms instantly.
It struggles with business logic, edge cases, and why we chose something.
Took 1000 hours to feel that difference. Now I know when to give 2 lines and when to give 2... | <start_of_turn>user
Write a LinkedIn post about: Why intuition for what AI grasps matters<end_of_turn>
<start_of_turn>model
My intuition for what Claude grasps instantly vs what needs more context changed everything.
It grasps boilerplate, patterns, and transforms instantly.
It struggles with business logic, edge cas... |
Write a LinkedIn post about: Context engineering vs prompt engineering | Prompt engineering is how you phrase it. Context engineering is what you give it.
I spent months tuning prompts. Gained maybe 10%.
Then I fixed the context - gave Claude the right files, the right history, the right constraints. Gained 50%.
Models don't need clever words. They need the right information.
Context is... | <start_of_turn>user
Write a LinkedIn post about: Context engineering vs prompt engineering<end_of_turn>
<start_of_turn>model
Prompt engineering is how you phrase it. Context engineering is what you give it.
I spent months tuning prompts. Gained maybe 10%.
Then I fixed the context - gave Claude the right files, the ri... |
Write a LinkedIn post about: Why I timebox everything | I timebox everything to 90 minutes.
If it doesn't fit, I cut scope, not time.
Constraints force clarity. Open-ended time invites over-engineering. | <start_of_turn>user
Write a LinkedIn post about: Why I timebox everything<end_of_turn>
<start_of_turn>model
I timebox everything to 90 minutes.
If it doesn't fit, I cut scope, not time.
Constraints force clarity. Open-ended time invites over-engineering.<end_of_turn>
|
Write a LinkedIn post about: Why I document decisions, not just code | Code tells you what. Docs should tell you why.
We started writing ADRs - 5 lines per decision. Why we chose X over Y, what we considered.
Claude reads them. So future sessions don't re-debate the same choice.
Without it, the agent re-invents the decision every time, slightly differently.
Document the why. The code ... | <start_of_turn>user
Write a LinkedIn post about: Why I document decisions, not just code<end_of_turn>
<start_of_turn>model
Code tells you what. Docs should tell you why.
We started writing ADRs - 5 lines per decision. Why we chose X over Y, what we considered.
Claude reads them. So future sessions don't re-debate the... |
Write a LinkedIn post about: What I do when I'm stuck | When I'm stuck, I do 3 things.
First, I write the problem in 3 sentences. Usually the stuck is vague thinking, not hard problem.
Second, I timebox to 30 minutes and try the dumbest solution. No cleverness.
Third, if still stuck, I ask Claude but with strict context: what I tried, what failed, what done looks like.
... | <start_of_turn>user
Write a LinkedIn post about: What I do when I'm stuck<end_of_turn>
<start_of_turn>model
When I'm stuck, I do 3 things.
First, I write the problem in 3 sentences. Usually the stuck is vague thinking, not hard problem.
Second, I timebox to 30 minutes and try the dumbest solution. No cleverness.
Thi... |
Write a LinkedIn post about: How I levelled up my prompts dramatically | My prompts leveled up when I stopped trying to be clever.
Old prompts were 300 words, with roles, constraints, and examples buried.
New prompts are 5 lines: context, task, constraint, example.
I moved the cleverness to CLAUDE.md. The prompt itself is boring.
Boring prompts are reliable prompts. The model doesn't ne... | <start_of_turn>user
Write a LinkedIn post about: How I levelled up my prompts dramatically<end_of_turn>
<start_of_turn>model
My prompts leveled up when I stopped trying to be clever.
Old prompts were 300 words, with roles, constraints, and examples buried.
New prompts are 5 lines: context, task, constraint, example.
... |
Write a LinkedIn post about: The second version is always better | The second version is always better. So ship the first fast.
You can't iterate on nothing.
v0 in a day, v1 in a week beats perfect in a month. | <start_of_turn>user
Write a LinkedIn post about: The second version is always better<end_of_turn>
<start_of_turn>model
The second version is always better. So ship the first fast.
You can't iterate on nothing.
v0 in a day, v1 in a week beats perfect in a month.<end_of_turn>
|
Write a LinkedIn post about: What I optimize for in 2025 | In 2025, I optimize for one thing: maintainability.
Not speed, not cleverness, not shipping count.
Can the codebase handle 5 more features without rewrite? Can a new person understand it in a day? Can Claude follow the patterns without guessing?
If yes, we are winning, even if we ship slower.
AI made building easy.... | <start_of_turn>user
Write a LinkedIn post about: What I optimize for in 2025<end_of_turn>
<start_of_turn>model
In 2025, I optimize for one thing: maintainability.
Not speed, not cleverness, not shipping count.
Can the codebase handle 5 more features without rewrite? Can a new person understand it in a day? Can Claude... |
Write a LinkedIn post about: My take on AI code maintenance | AI code is easy to write, harder to maintain. Same as human code, just faster.
The maintenance problem didn't change - code still needs to be readable, tested, and consistent.
What changed is volume. AI generates 10x more code. So maintenance matters 10x more.
I treat every AI diff as if a junior wrote it. I read it... | <start_of_turn>user
Write a LinkedIn post about: My take on AI code maintenance<end_of_turn>
<start_of_turn>model
AI code is easy to write, harder to maintain. Same as human code, just faster.
The maintenance problem didn't change - code still needs to be readable, tested, and consistent.
What changed is volume. AI g... |
Write a LinkedIn post about: How I write for scanners | People don't read. They scan.
So I write for scanners.
Short paragraphs (1-2 lines), one sentence per line for key points, no jargon, no big blocks.
Put the insight early. Don't bury it in paragraph 4.
I test by scrolling fast. If I can get the idea in 3 seconds, it's scannable. If not, I cut.
Optimize for the gla... | <start_of_turn>user
Write a LinkedIn post about: How I write for scanners<end_of_turn>
<start_of_turn>model
People don't read. They scan.
So I write for scanners.
Short paragraphs (1-2 lines), one sentence per line for key points, no jargon, no big blocks.
Put the insight early. Don't bury it in paragraph 4.
I test... |
Write a LinkedIn post about: Shipping a small project after 3 weeks of nights | Just shipped a small project I've been tinkering with for 3 weeks.
Nothing fancy. It writes LinkedIn posts that actually sound like me.
Most of the time was spent cutting words, not adding them. If a sentence didn't need to be there, it got deleted.
Try it and tell me what breaks. Link in comments. | <start_of_turn>user
Write a LinkedIn post about: Shipping a small project after 3 weeks of nights<end_of_turn>
<start_of_turn>model
Just shipped a small project I've been tinkering with for 3 weeks.
Nothing fancy. It writes LinkedIn posts that actually sound like me.
Most of the time was spent cutting words, not addi... |
Write a LinkedIn post about: How I pick what fundamentals to revisit | I know what fundamentals to revisit now. Because I saw where I got stuck.
When Claude handled something I didn't understand, I noted it. That's my study list.
How systems really work. How much context is enough to prompt well.
Not everything, just the gaps AI exposed.
AI shows you what you don't know. That's useful... | <start_of_turn>user
Write a LinkedIn post about: How I pick what fundamentals to revisit<end_of_turn>
<start_of_turn>model
I know what fundamentals to revisit now. Because I saw where I got stuck.
When Claude handled something I didn't understand, I noted it. That's my study list.
How systems really work. How much co... |
Write a LinkedIn post about: When to regenerate vs refactor | Should I refactor or regenerate?
If the file is small and messy, I regenerate. Faster, cleaner.
If the file is big and mostly good, I refactor. Preserve history, keep tests.
Rule of thumb: if I need to change more than 30% of the file, I prompt a fresh version with better context, then copy over.
Claude is great at... | <start_of_turn>user
Write a LinkedIn post about: When to regenerate vs refactor<end_of_turn>
<start_of_turn>model
Should I refactor or regenerate?
If the file is small and messy, I regenerate. Faster, cleaner.
If the file is big and mostly good, I refactor. Preserve history, keep tests.
Rule of thumb: if I need to c... |
Write a LinkedIn post about: Energy management > time management | Energy management > time management.
I code when energy is high. I do admin when it's low.
Matching task to energy beats any calendar hack. | <start_of_turn>user
Write a LinkedIn post about: Energy management > time management<end_of_turn>
<start_of_turn>model
Energy management > time management.
I code when energy is high. I do admin when it's low.
Matching task to energy beats any calendar hack.<end_of_turn>
|
Write a LinkedIn post about: Writing is thinking | Writing is thinking.
If your doc is confusing, your thinking is confusing. Fix the thinking, the doc fixes itself.
I write before I prompt. One paragraph: what I'm trying to do and why. Everything gets clearer. | <start_of_turn>user
Write a LinkedIn post about: Writing is thinking<end_of_turn>
<start_of_turn>model
Writing is thinking.
If your doc is confusing, your thinking is confusing. Fix the thinking, the doc fixes itself.
I write before I prompt. One paragraph: what I'm trying to do and why. Everything gets clearer.<end_... |
Write a LinkedIn post about: AI won't replace you, but someone using AI will | AI won't replace you.
Someone who can direct AI better than you will.
That's not about prompting. It's about knowing what good looks like, so you can tell when AI is wrong. | <start_of_turn>user
Write a LinkedIn post about: AI won't replace you, but someone using AI will<end_of_turn>
<start_of_turn>model
AI won't replace you.
Someone who can direct AI better than you will.
That's not about prompting. It's about knowing what good looks like, so you can tell when AI is wrong.<end_of_turn>
|
Write a LinkedIn post about: How I think about long-term consequences of autonomous choices | Every autonomous choice has a long-term cost.
File structure, naming, where logic lives. Model picks what works now.
I ask it to consider: if we build 5 more features on top of this, does this choice still make sense?
That one question filters out 70% of quick fixes.
Short-term thinking ships. Long-term thinking su... | <start_of_turn>user
Write a LinkedIn post about: How I think about long-term consequences of autonomous choices<end_of_turn>
<start_of_turn>model
Every autonomous choice has a long-term cost.
File structure, naming, where logic lives. Model picks what works now.
I ask it to consider: if we build 5 more features on to... |
Write a LinkedIn post about: Why I reply to every comment | I reply to every comment for the first hour. Every one.
Not with "thanks!" but with one real sentence.
Why? Comments are how posts spread, but also, it's just decent.
Someone took time to reply to you. Reply back like a human.
That one habit turned lurkers into collaborators. People remember who actually talks back... | <start_of_turn>user
Write a LinkedIn post about: Why I reply to every comment<end_of_turn>
<start_of_turn>model
I reply to every comment for the first hour. Every one.
Not with "thanks!" but with one real sentence.
Why? Comments are how posts spread, but also, it's just decent.
Someone took time to reply to you. Rep... |
Write a LinkedIn post about: How I end a post without a CTA | I don't end with "What do you think? 👇" anymore.
It feels needy. And it invites generic comments.
Now I just end. Or I ask a specific question to one person: "What's your capture system?"
Sometimes I just say "Peace ✌🏻" and leave. The post should earn replies, not beg for them.
If the idea is good, people reply w... | <start_of_turn>user
Write a LinkedIn post about: How I end a post without a CTA<end_of_turn>
<start_of_turn>model
I don't end with "What do you think? 👇" anymore.
It feels needy. And it invites generic comments.
Now I just end. Or I ask a specific question to one person: "What's your capture system?"
Sometimes I ju... |
Write a LinkedIn post about: Why I track decisions, not just tasks | We track decisions, not just tasks. Big difference.
Tasks say what we did. Decisions say why we chose it over alternatives.
We keep a decisions.md - 5 lines per decision. Date, context, choice, alternatives considered.
When a new session starts, Claude reads it. No re-debate, no drift.
Without it, every session re-... | <start_of_turn>user
Write a LinkedIn post about: Why I track decisions, not just tasks<end_of_turn>
<start_of_turn>model
We track decisions, not just tasks. Big difference.
Tasks say what we did. Decisions say why we chose it over alternatives.
We keep a decisions.md - 5 lines per decision. Date, context, choice, alt... |
Write a LinkedIn post about: The tradeoff between speed and quality I made | I used to pick speed. Always.
Shipped fast, fixed later. Later never came, and I had 10 fast messes.
Now I pick quality for core, speed for experiments.
Core: auth, payments, data - do it right, handle errors, test. Experiments: landing pages, scripts - ship fast, delete if fails.
Same team, two modes. Know which y... | <start_of_turn>user
Write a LinkedIn post about: The tradeoff between speed and quality I made<end_of_turn>
<start_of_turn>model
I used to pick speed. Always.
Shipped fast, fixed later. Later never came, and I had 10 fast messes.
Now I pick quality for core, speed for experiments.
Core: auth, payments, data - do it ... |
Write a LinkedIn post about: Building vs talking about building | Building vs talking about building. Very different activities.
Twitter rewards talking. Markets reward building.
I post after I ship, not before I start. | <start_of_turn>user
Write a LinkedIn post about: Building vs talking about building<end_of_turn>
<start_of_turn>model
Building vs talking about building. Very different activities.
Twitter rewards talking. Markets reward building.
I post after I ship, not before I start.<end_of_turn>
|
Write a LinkedIn post about: How I manage a big codebase with AI without losing control | Big codebase with AI feels like control is slipping. It is, if you let it.
How I keep it: every new file must state where it belongs and why. If Claude can't answer, I don't let it create the file.
Every change must reference the rulebook. No rule, no merge.
And one human reviews every AI diff before merge. No auto-... | <start_of_turn>user
Write a LinkedIn post about: How I manage a big codebase with AI without losing control<end_of_turn>
<start_of_turn>model
Big codebase with AI feels like control is slipping. It is, if you let it.
How I keep it: every new file must state where it belongs and why. If Claude can't answer, I don't let... |
Write a LinkedIn post about: Documentation is for future you | Documentation is for future you at 2am.
He has no memory of why you chose this. He will hate you if you didn't write it down.
5 lines in CLAUDE.md now saves 2 hours later. | <start_of_turn>user
Write a LinkedIn post about: Documentation is for future you<end_of_turn>
<start_of_turn>model
Documentation is for future you at 2am.
He has no memory of why you chose this. He will hate you if you didn't write it down.
5 lines in CLAUDE.md now saves 2 hours later.<end_of_turn>
|
Write a LinkedIn post about: The inspection paradox | The inspection paradox: the more you generate, the more you must inspect.
AI lets you create 10x. But you still have to read 10x.
If you can't inspect it, don't generate it. | <start_of_turn>user
Write a LinkedIn post about: The inspection paradox<end_of_turn>
<start_of_turn>model
The inspection paradox: the more you generate, the more you must inspect.
AI lets you create 10x. But you still have to read 10x.
If you can't inspect it, don't generate it.<end_of_turn>
|
Write a LinkedIn post about: The 1000 hour rule for AI tools | People want to be good at AI tools in a weekend. It takes longer.
I've used Claude Code daily for 6 months. Maybe 1000 hours. Only now do I feel like I have intuition.
Intuition for what it grasps instantly vs what needs more context. What will break, what will surprise.
There's no shortcut. You need the hours. The ... | <start_of_turn>user
Write a LinkedIn post about: The 1000 hour rule for AI tools<end_of_turn>
<start_of_turn>model
People want to be good at AI tools in a weekend. It takes longer.
I've used Claude Code daily for 6 months. Maybe 1000 hours. Only now do I feel like I have intuition.
Intuition for what it grasps instan... |
Write a LinkedIn post about: The quiet flex | The quiet flex is a codebase with no surprises.
No clever tricks. Just clear names, handled errors, obvious structure.
You notice it only when it's missing. | <start_of_turn>user
Write a LinkedIn post about: The quiet flex<end_of_turn>
<start_of_turn>model
The quiet flex is a codebase with no surprises.
No clever tricks. Just clear names, handled errors, obvious structure.
You notice it only when it's missing.<end_of_turn>
|
Write a LinkedIn post about: My simple tech stack for solo builders | My stack is boring. That's why it works.
Next.js, Postgres, Claude Code, Vercel. That's it.
No microservices. No custom infra. One repo.
If I need something, I prompt it first. If that fails, then I build.
Solo builders don't need scale. They need speed and low maintenance. Boring stack gives you both. | <start_of_turn>user
Write a LinkedIn post about: My simple tech stack for solo builders<end_of_turn>
<start_of_turn>model
My stack is boring. That's why it works.
Next.js, Postgres, Claude Code, Vercel. That's it.
No microservices. No custom infra. One repo.
If I need something, I prompt it first. If that fails, the... |
Write a LinkedIn post about: Why I prefer writing over meetings for decisions | We make big decisions in writing. Never in meetings.
Meeting notes are vague. Written proposals are precise.
We write a one-pager: context, options, tradeoffs, recommendation. Share async, comment, then meet only if stuck.
Takes more time upfront. Saves hours of re-debate.
Claude helps here too - it can turn messy ... | <start_of_turn>user
Write a LinkedIn post about: Why I prefer writing over meetings for decisions<end_of_turn>
<start_of_turn>model
We make big decisions in writing. Never in meetings.
Meeting notes are vague. Written proposals are precise.
We write a one-pager: context, options, tradeoffs, recommendation. Share asyn... |
Write a LinkedIn post about: How to actually learn anything | How to learn anything: build something tiny with it.
Not a course. Not a thread. One small project where it breaks.
The break teaches you more than the tutorial. | <start_of_turn>user
Write a LinkedIn post about: How to actually learn anything<end_of_turn>
<start_of_turn>model
How to learn anything: build something tiny with it.
Not a course. Not a thread. One small project where it breaks.
The break teaches you more than the tutorial.<end_of_turn>
|
Write a LinkedIn post about: The mindset win from seeing anything can be figured out | The biggest win from Kiro wasn't productivity. It was mindset.
I used to think some things were too hard for me. Now I've seen that if you sit calmly and iterate, anything can be figured out.
That's not AI magic. It's proof that patient iteration works.
You carry that mindset even without the tool.
Given breathing ... | <start_of_turn>user
Write a LinkedIn post about: The mindset win from seeing anything can be figured out<end_of_turn>
<start_of_turn>model
The biggest win from Kiro wasn't productivity. It was mindset.
I used to think some things were too hard for me. Now I've seen that if you sit calmly and iterate, anything can be f... |
Write a LinkedIn post about: The telltale signs of AI-written LinkedIn | You can spot AI LinkedIn in 2 seconds.
Em dashes everywhere — like this —, "delve", "tapestry", "unlock the power", and that same 5-paragraph structure.
Also, too perfect grammar, too balanced take, too many emojis.
My filter: read it and ask, would I say this to a friend? If not, it's AI.
Simple test, but catches ... | <start_of_turn>user
Write a LinkedIn post about: The telltale signs of AI-written LinkedIn<end_of_turn>
<start_of_turn>model
You can spot AI LinkedIn in 2 seconds.
Em dashes everywhere — like this —, "delve", "tapestry", "unlock the power", and that same 5-paragraph structure.
Also, too perfect grammar, too balanced ... |
Write a LinkedIn post about: The side project that taught me more than my job | I built a throwaway tool last month. 2 days, no users, just for me.
It auto-sorts my downloads folder. Dumb problem, dumb solution. A Claude Code skill that runs daily.
But it taught me 3 things my job didn't.
First, how to scope ruthlessly. No auth, no UI, just a script. Second, how to prompt for disposable code - ... | <start_of_turn>user
Write a LinkedIn post about: The side project that taught me more than my job<end_of_turn>
<start_of_turn>model
I built a throwaway tool last month. 2 days, no users, just for me.
It auto-sorts my downloads folder. Dumb problem, dumb solution. A Claude Code skill that runs daily.
But it taught me ... |
Write a LinkedIn post about: You don't need more ideas | You don't need more ideas. You need to execute one.
Ideas are cheap. Twitter has 100 per hour.
Pick the smallest one you can ship in a week. Ship it. Learn. Repeat. | <start_of_turn>user
Write a LinkedIn post about: You don't need more ideas<end_of_turn>
<start_of_turn>model
You don't need more ideas. You need to execute one.
Ideas are cheap. Twitter has 100 per hour.
Pick the smallest one you can ship in a week. Ship it. Learn. Repeat.<end_of_turn>
|
Write a LinkedIn post about: The myth of the perfect stack | People debate stack like it's destiny. It's not.
Pick boring, pick familiar, pick what you can debug at 2am.
Our stack is Next.js, Postgres, Vercel. Not because it's best, but because we know where it breaks.
Perfect stack is a myth. Comfortable stack is real.
Spend stack debate time on shipping. The stack won't sa... | <start_of_turn>user
Write a LinkedIn post about: The myth of the perfect stack<end_of_turn>
<start_of_turn>model
People debate stack like it's destiny. It's not.
Pick boring, pick familiar, pick what you can debug at 2am.
Our stack is Next.js, Postgres, Vercel. Not because it's best, but because we know where it brea... |
Write a LinkedIn post about: How I debug production issues now | Production broke at 4pm. Old me would have panicked.
New me has a checklist. And checklist beats panic every time.
Step 1: what changed? Check last 5 commits, last deploy. Step 2: who is affected? One user or everyone? Step 3: can I revert? If yes, revert first, debug next.
This time it was a DB migration that locke... | <start_of_turn>user
Write a LinkedIn post about: How I debug production issues now<end_of_turn>
<start_of_turn>model
Production broke at 4pm. Old me would have panicked.
New me has a checklist. And checklist beats panic every time.
Step 1: what changed? Check last 5 commits, last deploy. Step 2: who is affected? One ... |
Write a LinkedIn post about: Reading code > writing code | Reading code > writing code.
I spend 70% of my time reading. 30% writing.
AI flipped the ratio for others - they write 90%, read 10%. That's how bugs are born. | <start_of_turn>user
Write a LinkedIn post about: Reading code > writing code<end_of_turn>
<start_of_turn>model
Reading code > writing code.
I spend 70% of my time reading. 30% writing.
AI flipped the ratio for others - they write 90%, read 10%. That's how bugs are born.<end_of_turn>
|
Write a LinkedIn post about: Why naming matters more than clever code | Clever code feels smart for a day. Good naming pays for months.
I've deleted more clever code than I've kept. But good names stay.
If a function name needs a comment, rename it.
I tell Claude the same: use full words, no abbreviations, name should tell what it does.
Boring names, readable codebase. | <start_of_turn>user
Write a LinkedIn post about: Why naming matters more than clever code<end_of_turn>
<start_of_turn>model
Clever code feels smart for a day. Good naming pays for months.
I've deleted more clever code than I've kept. But good names stay.
If a function name needs a comment, rename it.
I tell Claude t... |
Write a LinkedIn post about: The art of deleting features | Deleting features is harder than adding them. But better.
We removed a whole analytics page last month. 3 weeks of work, gone.
Nobody complained. One user said "oh, I didn't know that existed."
Product got lighter, faster, clearer. Team felt relief, not loss.
Every feature has a maintenance cost. If usage doesn't j... | <start_of_turn>user
Write a LinkedIn post about: The art of deleting features<end_of_turn>
<start_of_turn>model
Deleting features is harder than adding them. But better.
We removed a whole analytics page last month. 3 weeks of work, gone.
Nobody complained. One user said "oh, I didn't know that existed."
Product got... |
Write a LinkedIn post about: Why I prefer disposable scripts over full apps for personal tools | I used to turn every personal problem into a full project.
Domain, auth, deployment. For a tool only I would use.
Now I just write a Claude Code skill. Or a 50 line script I can regenerate tomorrow.
Skips 80% of the decisions. Zero maintenance.
If only you need it and you need it once, don't build a product. Build ... | <start_of_turn>user
Write a LinkedIn post about: Why I prefer disposable scripts over full apps for personal tools<end_of_turn>
<start_of_turn>model
I used to turn every personal problem into a full project.
Domain, auth, deployment. For a tool only I would use.
Now I just write a Claude Code skill. Or a 50 line scri... |
Write a LinkedIn post about: Learning AWS by breaking things | I learned AWS by breaking it. Repeatedly.
S3 bucket public when it shouldn't be. Security group open to 0.0.0.0. Lambda timeout at 3 seconds.
Each break taught me more than a course.
Now I keep a checklist in CLAUDE.md: what to check before any AWS change. Claude follows it, so I don't repeat mistakes.
Break small,... | <start_of_turn>user
Write a LinkedIn post about: Learning AWS by breaking things<end_of_turn>
<start_of_turn>model
I learned AWS by breaking it. Repeatedly.
S3 bucket public when it shouldn't be. Security group open to 0.0.0.0. Lambda timeout at 3 seconds.
Each break taught me more than a course.
Now I keep a checkl... |
Write a LinkedIn post about: How I decide to build vs buy vs prompt | My decision tree is simple now.
Can I prompt it in 10 minutes? Prompt it.
Can I buy it for $20/month? Buy it.
Otherwise, build it.
Prompt is my new first option. It's disposable, fast, and good enough for 60% of personal tooling.
Only build if it needs to persist, scale, or be shared.
Most things don't. | <start_of_turn>user
Write a LinkedIn post about: How I decide to build vs buy vs prompt<end_of_turn>
<start_of_turn>model
My decision tree is simple now.
Can I prompt it in 10 minutes? Prompt it.
Can I buy it for $20/month? Buy it.
Otherwise, build it.
Prompt is my new first option. It's disposable, fast, and good en... |
Write a LinkedIn post about: Speed is a feature | Speed is a feature.
Users feel 200ms. They don't feel your clean architecture.
Make it fast first, clean later. Nobody praises maintainable but slow. | <start_of_turn>user
Write a LinkedIn post about: Speed is a feature<end_of_turn>
<start_of_turn>model
Speed is a feature.
Users feel 200ms. They don't feel your clean architecture.
Make it fast first, clean later. Nobody praises maintainable but slow.<end_of_turn>
|
Write a LinkedIn post about: Productivity boost from AI during crunch time | Past 2 months were crunch. AI made it survivable.
Not by writing everything, but by handling the repetitive 60%. Boilerplate, tests, docs.
I saved brain for the hard 40% - decisions, design, debugging.
Productivity boost was clutch, not magic. It gave me space to think.
Now with breathing room, I'm using that space... | <start_of_turn>user
Write a LinkedIn post about: Productivity boost from AI during crunch time<end_of_turn>
<start_of_turn>model
Past 2 months were crunch. AI made it survivable.
Not by writing everything, but by handling the repetitive 60%. Boilerplate, tests, docs.
I saved brain for the hard 40% - decisions, design... |
Write a LinkedIn post about: My experience with Kiro, the AI-powered code editor from AWS | This was long overdue, and I'm writing it today.
Kiro is the greatest piece of software made by mankind.
I was lucky enough to use it from day one (can't thank Hardik Shah enough for the early access). For those who don't know, Kiro is an AI-powered code editor backed by AWS itself: agentic and game-changer (just lik... | <start_of_turn>user
Write a LinkedIn post about: My experience with Kiro, the AI-powered code editor from AWS<end_of_turn>
<start_of_turn>model
This was long overdue, and I'm writing it today.
Kiro is the greatest piece of software made by mankind.
I was lucky enough to use it from day one (can't thank Hardik Shah en... |
Write a LinkedIn post about: Knowing when to lean on AI and when not to | I lean on AI for speed. I don't lean on it for judgment.
AI writes the first draft. I decide if it's the right draft.
During crunch, I leaned hard. Now I'm leaning less, learning more.
The downsides of leaning too hard are real - you stop learning the fundamentals.
Balance is knowing when to delegate and when to do... | <start_of_turn>user
Write a LinkedIn post about: Knowing when to lean on AI and when not to<end_of_turn>
<start_of_turn>model
I lean on AI for speed. I don't lean on it for judgment.
AI writes the first draft. I decide if it's the right draft.
During crunch, I leaned hard. Now I'm leaning less, learning more.
The do... |
Write a LinkedIn post about: Saying no to a feature that didn't need to exist | We almost built it. Dashboard, filters, export. Two weeks of work.
Then I asked: who actually needs this?
Turned out one person asked once. We had a spreadsheet that already solved it.
So we said no. Wrote a 20 line script instead.
Best feature is the one you don't build. More true with AI now, when building feels ... | <start_of_turn>user
Write a LinkedIn post about: Saying no to a feature that didn't need to exist<end_of_turn>
<start_of_turn>model
We almost built it. Dashboard, filters, export. Two weeks of work.
Then I asked: who actually needs this?
Turned out one person asked once. We had a spreadsheet that already solved it.
... |
Write a LinkedIn post about: Building trust by sounding normal | Trust comes from sounding normal.
No hype, no "game-changer", no rocket emojis. Just honest: "We built this, here's what broke, here's what we learned."
When I sound like a human talking to a friend, people DM me.
When I sounded like a LinkedIn guru, people scrolled.
Normal is a competitive advantage now. Most cont... | <start_of_turn>user
Write a LinkedIn post about: Building trust by sounding normal<end_of_turn>
<start_of_turn>model
Trust comes from sounding normal.
No hype, no "game-changer", no rocket emojis. Just honest: "We built this, here's what broke, here's what we learned."
When I sound like a human talking to a friend, p... |
Write a LinkedIn post about: The difference between completing a task and building a foundation | Completing a task is easy. Building a foundation is hard.
Task: add login. Foundation: add auth that next 5 features can reuse without copy-paste.
Models are great at tasks. They need help with foundations.
I now review every task with: can the next feature reuse this? If not, redo it.
Tasks ship. Foundations last. | <start_of_turn>user
Write a LinkedIn post about: The difference between completing a task and building a foundation<end_of_turn>
<start_of_turn>model
Completing a task is easy. Building a foundation is hard.
Task: add login. Foundation: add auth that next 5 features can reuse without copy-paste.
Models are great at t... |
Write a LinkedIn post about: Why fundamentals still win long term | Fundamentals still win long term. Even with AI.
Everyone learns the new model, new tool. Few revisit how systems actually work.
But fundamentals are what let you judge AI output. Without them, you can't tell when it's wrong.
I'm circling back to basics now — how DBs work, how networks work, how to reason.
Not sexy,... | <start_of_turn>user
Write a LinkedIn post about: Why fundamentals still win long term<end_of_turn>
<start_of_turn>model
Fundamentals still win long term. Even with AI.
Everyone learns the new model, new tool. Few revisit how systems actually work.
But fundamentals are what let you judge AI output. Without them, you c... |
Write a LinkedIn post about: Small team, big output | Small team, big output is the new flex.
3 people with Claude can outship 10 without it. But only if they have good taste and clear rules.
Headcount is vanity. Output is sanity. | <start_of_turn>user
Write a LinkedIn post about: Small team, big output<end_of_turn>
<start_of_turn>model
Small team, big output is the new flex.
3 people with Claude can outship 10 without it. But only if they have good taste and clear rules.
Headcount is vanity. Output is sanity.<end_of_turn>
|
Write a LinkedIn post about: Goodhart's law for metrics | When a metric becomes a target, it ceases to be a good metric.
We stopped optimizing for lines of code generated. Started optimizing for bugs not created.
Everything changed. | <start_of_turn>user
Write a LinkedIn post about: Goodhart's law for metrics<end_of_turn>
<start_of_turn>model
When a metric becomes a target, it ceases to be a good metric.
We stopped optimizing for lines of code generated. Started optimizing for bugs not created.
Everything changed.<end_of_turn>
|
Write a LinkedIn post about: Why one idea per post works | One idea per post. That's my rule.
Not three learnings, not a thread disguised as a post. One idea, explained simply.
Why? People scroll fast. They remember one thing, not five.
When I tried to pack 3 ideas, comments were confused. When I pick one, replies are precise.
Constraint helps. "What's the single takeaway?... | <start_of_turn>user
Write a LinkedIn post about: Why one idea per post works<end_of_turn>
<start_of_turn>model
One idea per post. That's my rule.
Not three learnings, not a thread disguised as a post. One idea, explained simply.
Why? People scroll fast. They remember one thing, not five.
When I tried to pack 3 ideas... |
Write a LinkedIn post about: The trap of productivity tools | The trap of productivity tools: you spend more time optimizing them than doing work.
I use one doc, one board, one chat. Everything else is distraction dressed as productivity. | <start_of_turn>user
Write a LinkedIn post about: The trap of productivity tools<end_of_turn>
<start_of_turn>model
The trap of productivity tools: you spend more time optimizing them than doing work.
I use one doc, one board, one chat. Everything else is distraction dressed as productivity.<end_of_turn>
|
Write a LinkedIn post about: The best advice I got from Hardik Shah | Best advice I got from Hardik Shah: get early access, then actually use it.
He gave me Kiro on day one. Most people would wait, watch tutorials.
I just built with it. Broke things, fixed them, learned what it could actually do.
That hands-on time taught me more than any demo.
Access means nothing without iteration. | <start_of_turn>user
Write a LinkedIn post about: The best advice I got from Hardik Shah<end_of_turn>
<start_of_turn>model
Best advice I got from Hardik Shah: get early access, then actually use it.
He gave me Kiro on day one. Most people would wait, watch tutorials.
I just built with it. Broke things, fixed them, lea... |
Write a LinkedIn post about: How we cut our LLM costs by 70% | Our LLM bill was scary. $800/month for internal tools.
Cut it by 70% without losing quality.
First, we cached. Same prompt with same context? Don't call the model. Second, we used smaller models for smaller tasks - Haiku for simple rewrites, Sonnet for planning. Third, we put hard limits in prompts: "max 150 words" s... | <start_of_turn>user
Write a LinkedIn post about: How we cut our LLM costs by 70%<end_of_turn>
<start_of_turn>model
Our LLM bill was scary. $800/month for internal tools.
Cut it by 70% without losing quality.
First, we cached. Same prompt with same context? Don't call the model. Second, we used smaller models for smal... |
Write a LinkedIn post about: What I look for in a code review | My code review checklist fits on a sticky note.
1. Can I understand this without asking the author?
2. Will this make the next feature easier or harder?
3. Are errors explicit?
4. Is there a test for the sad path?
That's it.
I don't care about cleverness. I care about whether the code respects the next person who re... | <start_of_turn>user
Write a LinkedIn post about: What I look for in a code review<end_of_turn>
<start_of_turn>model
My code review checklist fits on a sticky note.
1. Can I understand this without asking the author?
2. Will this make the next feature easier or harder?
3. Are errors explicit?
4. Is there a test for the... |
Write a LinkedIn post about: A bug that took 2 days to find and 2 lines to fix | 2 days. That's how long I chased this bug.
Auth worked locally. Failed in staging. Logs looked fine. Tests passed.
Turned out to be a trailing slash. The API expected `/api/user` and I was sending `/api/user/`. Nginx silently redirected, dropped the auth header.
Fix was 2 lines. In the .conf file that is literally o... | <start_of_turn>user
Write a LinkedIn post about: A bug that took 2 days to find and 2 lines to fix<end_of_turn>
<start_of_turn>model
2 days. That's how long I chased this bug.
Auth worked locally. Failed in staging. Logs looked fine. Tests passed.
Turned out to be a trailing slash. The API expected `/api/user` and I ... |
Write a LinkedIn post about: Iteration over perfection | I used to aim for perfect on first try. Spent days planning.
Now I ship v0 in an hour. Let Claude generate it, then iterate 5 times.
Each iteration is calmer, more focused. Literally anything can be figured out if you sit down and iterate.
Perfection is a guess. Iteration is learning.
You get to good faster by bein... | <start_of_turn>user
Write a LinkedIn post about: Iteration over perfection<end_of_turn>
<start_of_turn>model
I used to aim for perfect on first try. Spent days planning.
Now I ship v0 in an hour. Let Claude generate it, then iterate 5 times.
Each iteration is calmer, more focused. Literally anything can be figured ou... |
Write a LinkedIn post about: My 10 min post workflow | My post workflow is 10 minutes.
2 mins: pick topic from topics.txt, write one sentence: what's the takeaway?
3 mins: draft, messy, no editing.
5 mins: cut. Shorten sentences, delete fluff, make it sound like me talking.
No design, no hashtags, no perfect image.
Posting consistently beats posting perfectly. 10 minute... | <start_of_turn>user
Write a LinkedIn post about: My 10 min post workflow<end_of_turn>
<start_of_turn>model
My post workflow is 10 minutes.
2 mins: pick topic from topics.txt, write one sentence: what's the takeaway?
3 mins: draft, messy, no editing.
5 mins: cut. Shorten sentences, delete fluff, make it sound like me t... |
Write a LinkedIn post about: The loneliness of solo building and how I handle it | Solo building is lonely. Nobody tells you that.
No standup, no review, just you and Claude. Some days it feels like shouting into void.
What helps: posting progress, even small. A tweet, a LinkedIn post, a DM to a friend. External feedback breaks the loop.
Second: have one human you talk to weekly about the work. No... | <start_of_turn>user
Write a LinkedIn post about: The loneliness of solo building and how I handle it<end_of_turn>
<start_of_turn>model
Solo building is lonely. Nobody tells you that.
No standup, no review, just you and Claude. Some days it feels like shouting into void.
What helps: posting progress, even small. A twe... |
Write a LinkedIn post about: How I keep codebase flexible for next feature | I keep codebase flexible with one question: where will the next feature go?
Before writing code, I point to the file, the function, the reuse path.
If I can't point, the structure is wrong.
Tell Claude the same: before you write, say where the next feature plugs in.
Flexible code isn't accident. It's planned reuse. | <start_of_turn>user
Write a LinkedIn post about: How I keep codebase flexible for next feature<end_of_turn>
<start_of_turn>model
I keep codebase flexible with one question: where will the next feature go?
Before writing code, I point to the file, the function, the reuse path.
If I can't point, the structure is wrong.... |
Write a LinkedIn post about: How I use AI without sounding like AI | People ask how I use AI without sounding like AI.
Simple: I never publish AI output directly.
AI gives me draft, I rewrite in my voice. Shorten sentences, add the bracket aside, cut the fluff.
My filter: would I say this to a friend over coffee? If not, rewrite.
Also, I keep a list of AI tells: "delve", "tapestry",... | <start_of_turn>user
Write a LinkedIn post about: How I use AI without sounding like AI<end_of_turn>
<start_of_turn>model
People ask how I use AI without sounding like AI.
Simple: I never publish AI output directly.
AI gives me draft, I rewrite in my voice. Shorten sentences, add the bracket aside, cut the fluff.
My ... |
Write a LinkedIn post about: What I wish I knew at 22 | What I wish I knew at 22: shipping teaches faster than studying.
I spent years preparing to build. Should have just built badly and learned.
Permission comes after you start, not before. | <start_of_turn>user
Write a LinkedIn post about: What I wish I knew at 22<end_of_turn>
<start_of_turn>model
What I wish I knew at 22: shipping teaches faster than studying.
I spent years preparing to build. Should have just built badly and learned.
Permission comes after you start, not before.<end_of_turn>
|
Write a LinkedIn post about: How I learned to say no to investors | We said no to money. It was hard.
Investor wanted us to add enterprise features. Big check, big roadmap change.
But we wanted to build for solo builders, not enterprises. Different product, different soul.
Saying yes would have meant hiring, process, and building for someone else's needs.
So we said no. Kept buildi... | <start_of_turn>user
Write a LinkedIn post about: How I learned to say no to investors<end_of_turn>
<start_of_turn>model
We said no to money. It was hard.
Investor wanted us to add enterprise features. Big check, big roadmap change.
But we wanted to build for solo builders, not enterprises. Different product, differen... |
Write a LinkedIn post about: Why I keep a changelog for myself | I keep a changelog for myself. Not for users, for me.
Every Friday, 5 bullets: what shipped, what broke, what I learned.
Over months, it shows patterns. I break the same thing, I avoid the same learning.
Also, on low days, I read it. Proof that I'm moving.
Claude updates it now too - reads my commits, drafts the bu... | <start_of_turn>user
Write a LinkedIn post about: Why I keep a changelog for myself<end_of_turn>
<start_of_turn>model
I keep a changelog for myself. Not for users, for me.
Every Friday, 5 bullets: what shipped, what broke, what I learned.
Over months, it shows patterns. I break the same thing, I avoid the same learnin... |
Write a LinkedIn post about: Why I stopped using JIRA | We stopped using JIRA. Productivity went up.
Not because JIRA is bad. Because we were using it to avoid talking about priorities.
20 tickets in "In Progress" means nothing is actually in progress.
Now we have a markdown file. 3 columns: Now, Next, Later. Max 2 things in Now.
Claude reads it, so it knows what matter... | <start_of_turn>user
Write a LinkedIn post about: Why I stopped using JIRA<end_of_turn>
<start_of_turn>model
We stopped using JIRA. Productivity went up.
Not because JIRA is bad. Because we were using it to avoid talking about priorities.
20 tickets in "In Progress" means nothing is actually in progress.
Now we have ... |
Write a LinkedIn post about: How I validate an idea in 48 hours | Validate in 48 hours or don't validate at all.
My method: post the idea on LinkedIn/Twitter, build a landing page with a waitlist in 2 hours. If 10 people sign up, build. If not, kill.
No code before signal.
I killed 4 ideas last month this way. Saved months of building.
One got 30 signups. Built it in a week. Peop... | <start_of_turn>user
Write a LinkedIn post about: How I validate an idea in 48 hours<end_of_turn>
<start_of_turn>model
Validate in 48 hours or don't validate at all.
My method: post the idea on LinkedIn/Twitter, build a landing page with a waitlist in 2 hours. If 10 people sign up, build. If not, kill.
No code before ... |
Write a LinkedIn post about: The 1% better every day lie | 1% better every day sounds nice. It's a lie.
Growth is spikes and plateaus, not a smooth curve.
Some weeks you leap, some weeks you crawl. Both are part of it. | <start_of_turn>user
Write a LinkedIn post about: The 1% better every day lie<end_of_turn>
<start_of_turn>model
1% better every day sounds nice. It's a lie.
Growth is spikes and plateaus, not a smooth curve.
Some weeks you leap, some weeks you crawl. Both are part of it.<end_of_turn>
|
Write a LinkedIn post about: What I learned after 30 days of daily writing | Wrote for 30 days straight. Posted every day, even when it was bad. Especially when it was bad.
Week 1: tried to sound smart. Got no engagement, felt fake.
Week 2: tried to be viral. Wrote cringe, deleted half.
Week 3: stopped trying. Wrote like I talk. Short sentences, one idea, honest. People started replying.
The ... | <start_of_turn>user
Write a LinkedIn post about: What I learned after 30 days of daily writing<end_of_turn>
<start_of_turn>model
Wrote for 30 days straight. Posted every day, even when it was bad. Especially when it was bad.
Week 1: tried to sound smart. Got no engagement, felt fake.
Week 2: tried to be viral. Wrote c... |
Write a LinkedIn post about: Why I write for one person | I write for one person. Not an audience.
Imagine one friend who gets it, but needs the nudge. Write to him.
Audience writing makes you vague and cringe - "excited to share", "humbled". One-person writing makes you direct.
My best posts were written to Hardik. One person, one idea, one nudge.
When you write for ever... | <start_of_turn>user
Write a LinkedIn post about: Why I write for one person<end_of_turn>
<start_of_turn>model
I write for one person. Not an audience.
Imagine one friend who gets it, but needs the nudge. Write to him.
Audience writing makes you vague and cringe - "excited to share", "humbled". One-person writing make... |
Write a LinkedIn post about: Taking a break from ongoing projects to revisit fundamentals | Taking a break from shipping to study fundamentals feels slow. It's not.
I paused new features for 2 weeks. Just read how systems actually work - DB indexes, network, OS basics.
Came back and my prompts got sharper. I knew what to ask, what to skip, what the model was getting wrong.
You need to lean on AI, but you c... | <start_of_turn>user
Write a LinkedIn post about: Taking a break from ongoing projects to revisit fundamentals<end_of_turn>
<start_of_turn>model
Taking a break from shipping to study fundamentals feels slow. It's not.
I paused new features for 2 weeks. Just read how systems actually work - DB indexes, network, OS basic... |
Write a LinkedIn post about: The curse of the clever solution | The curse of the clever solution: you understand it today, nobody understands it next month.
Choose the boring solution. It's clever in a different way - it's obvious. | <start_of_turn>user
Write a LinkedIn post about: The curse of the clever solution<end_of_turn>
<start_of_turn>model
The curse of the clever solution: you understand it today, nobody understands it next month.
Choose the boring solution. It's clever in a different way - it's obvious.<end_of_turn>
|
Write a LinkedIn post about: The real cost of a messy codebase | Messy codebase doesn't cost you today. It costs you every day after.
We measured it. One messy module added 30 minutes to every feature that touched it. 10 features later, that's 5 hours.
That's a whole rulebook worth of time, just lost to hesitation.
We cleaned it in a day. Added error contracts, consistent naming,... | <start_of_turn>user
Write a LinkedIn post about: The real cost of a messy codebase<end_of_turn>
<start_of_turn>model
Messy codebase doesn't cost you today. It costs you every day after.
We measured it. One messy module added 30 minutes to every feature that touched it. 10 features later, that's 5 hours.
That's a whol... |
Write a LinkedIn post about: My weekly reflection format | Every Friday I do a 15 minute reflection. 3 sections.
What shipped: 3 bullets, what actually went live.
What broke: 1-2 bullets, what failed and why.
What I learned: 1 bullet, the single lesson.
That's it. Takes 15 minutes, saves hours of context loss.
Claude drafts it from my commits, I edit. Future me loves it.
S... | <start_of_turn>user
Write a LinkedIn post about: My weekly reflection format<end_of_turn>
<start_of_turn>model
Every Friday I do a 15 minute reflection. 3 sections.
What shipped: 3 bullets, what actually went live.
What broke: 1-2 bullets, what failed and why.
What I learned: 1 bullet, the single lesson.
That's it. T... |
Write a LinkedIn post about: Why I delete more than I write | I delete more than I write now.
Every feature makes the product heavier.
The best PM I know has a delete key mentality. Subtraction is progress. | <start_of_turn>user
Write a LinkedIn post about: Why I delete more than I write<end_of_turn>
<start_of_turn>model
I delete more than I write now.
Every feature makes the product heavier.
The best PM I know has a delete key mentality. Subtraction is progress.<end_of_turn>
|
Write a LinkedIn post about: Delegating to AI vs delegating to humans - similarities | Delegating to AI feels a lot like delegating to humans.
You need to give context, define done, and check the work.
If you're vague, both will fill in gaps with their best guess. Both will be confidently wrong.
Difference is, humans will ask clarifying questions. AI will just proceed.
So you have to be more explicit... | <start_of_turn>user
Write a LinkedIn post about: Delegating to AI vs delegating to humans - similarities<end_of_turn>
<start_of_turn>model
Delegating to AI feels a lot like delegating to humans.
You need to give context, define done, and check the work.
If you're vague, both will fill in gaps with their best guess. B... |
Write a LinkedIn post about: Taste is the new skill | In the AI era, taste is the skill.
Everyone can generate. Few can judge what's good.
Taste comes from building, reading good work, and throwing away bad work. No shortcut. | <start_of_turn>user
Write a LinkedIn post about: Taste is the new skill<end_of_turn>
<start_of_turn>model
In the AI era, taste is the skill.
Everyone can generate. Few can judge what's good.
Taste comes from building, reading good work, and throwing away bad work. No shortcut.<end_of_turn>
|
Write a LinkedIn post about: Why basics still matter in 2025 | Everyone chases the new model. I went back to basics.
Naming, structure, error handling, testing. Nothing groundbreaking for a software engineer.
But basics are what the model gets wrong when left alone.
And basics are what makes AI output maintainable.
In 2025, basics are competitive advantage. | <start_of_turn>user
Write a LinkedIn post about: Why basics still matter in 2025<end_of_turn>
<start_of_turn>model
Everyone chases the new model. I went back to basics.
Naming, structure, error handling, testing. Nothing groundbreaking for a software engineer.
But basics are what the model gets wrong when left alone.... |
Write a LinkedIn post about: Focus is saying no 100 times | Focus is saying no 100 times.
We cut 5 features last month. Feels bad. Product got better.
Focus isn't what you say yes to. It's what you have the guts to say no to. | <start_of_turn>user
Write a LinkedIn post about: Focus is saying no 100 times<end_of_turn>
<start_of_turn>model
Focus is saying no 100 times.
We cut 5 features last month. Feels bad. Product got better.
Focus isn't what you say yes to. It's what you have the guts to say no to.<end_of_turn>
|
Write a LinkedIn post about: How I beat procrastination | How I beat procrastination: make the first step 2 minutes.
Not "build feature". Just "open file, write function signature".
Momentum does the rest. Starting is the only hard part. | <start_of_turn>user
Write a LinkedIn post about: How I beat procrastination<end_of_turn>
<start_of_turn>model
How I beat procrastination: make the first step 2 minutes.
Not "build feature". Just "open file, write function signature".
Momentum does the rest. Starting is the only hard part.<end_of_turn>
|
Write a LinkedIn post about: Abstractions that leaked and how I fixed them | Our payment abstraction leaked. Badly.
It hid whether we were calling Stripe or a mock. Tests passed, production failed on refunds.
We had abstracted too early, before we understood the difference.
Fix was to remove the abstraction. Just call Stripe directly, handle the cases plainly.
Add abstraction when you have ... | <start_of_turn>user
Write a LinkedIn post about: Abstractions that leaked and how I fixed them<end_of_turn>
<start_of_turn>model
Our payment abstraction leaked. Badly.
It hid whether we were calling Stripe or a mock. Tests passed, production failed on refunds.
We had abstracted too early, before we understood the dif... |
Write a LinkedIn post about: Posting consistently without burning out | Posting consistently without burning out is about scope, not discipline.
I don't try to write viral posts. I try to write one honest 60-word post.
That's small enough to do even on low days.
Some days it's a tweet-like thought. Some days it's a longer story. Both count.
Consistency comes from making the bar so low ... | <start_of_turn>user
Write a LinkedIn post about: Posting consistently without burning out<end_of_turn>
<start_of_turn>model
Posting consistently without burning out is about scope, not discipline.
I don't try to write viral posts. I try to write one honest 60-word post.
That's small enough to do even on low days.
So... |
Write a LinkedIn post about: Shipping fast vs shipping right | I used to ship fast. Now I ship slightly slower, but I sleep better.
Fast meant copy-paste, no tests, "we'll fix it later". Later never came.
Right doesn't mean perfect. It means: named well, error handled, next feature can reuse it.
Adds maybe 20% time. Saves 50% in maintenance.
Still ship fast for disposable stuf... | <start_of_turn>user
Write a LinkedIn post about: Shipping fast vs shipping right<end_of_turn>
<start_of_turn>model
I used to ship fast. Now I ship slightly slower, but I sleep better.
Fast meant copy-paste, no tests, "we'll fix it later". Later never came.
Right doesn't mean perfect. It means: named well, error handl... |
Write a LinkedIn post about: Why I stopped chasing the newest model | I stopped chasing the newest model. Kept using Sonnet while everyone hyped the next thing.
New model drops, everyone benchmarks, switches prompts, re-learns quirks.
I stayed, deepened my context, improved my rulebook. Got more out of the old model than they got out of the new.
Chasing models is a treadmill. Deepenin... | <start_of_turn>user
Write a LinkedIn post about: Why I stopped chasing the newest model<end_of_turn>
<start_of_turn>model
I stopped chasing the newest model. Kept using Sonnet while everyone hyped the next thing.
New model drops, everyone benchmarks, switches prompts, re-learns quirks.
I stayed, deepened my context, ... |
Write a LinkedIn post about: Rewriting a codebase at 2am - was it worth it | Rewrote the whole service at 2am. Felt productive.
Morning came. Nothing worked. Tests failed, types broke, Claude had been confidently wrong for 3 hours.
Agentic coding at 2am is still coding at 2am. Your judgment is off, and the model won't tell you.
I reverted. Did the same rewrite at 10am in 2 hours, properly.
... | <start_of_turn>user
Write a LinkedIn post about: Rewriting a codebase at 2am - was it worth it<end_of_turn>
<start_of_turn>model
Rewrote the whole service at 2am. Felt productive.
Morning came. Nothing worked. Tests failed, types broke, Claude had been confidently wrong for 3 hours.
Agentic coding at 2am is still cod... |
Write a LinkedIn post about: Build for one, not for everyone | Build for one person. Not for everyone.
When you build for everyone, you build for no one.
My best tools were built for me, then 3 friends wanted them. That's signal. | <start_of_turn>user
Write a LinkedIn post about: Build for one, not for everyone<end_of_turn>
<start_of_turn>model
Build for one person. Not for everyone.
When you build for everyone, you build for no one.
My best tools were built for me, then 3 friends wanted them. That's signal.<end_of_turn>
|
Write a LinkedIn post about: The best code is no code | Best code is no code.
Second best is code you can delete tomorrow.
I check this before every feature: can I solve this with a prompt instead of a codebase? Half the time, yes. | <start_of_turn>user
Write a LinkedIn post about: The best code is no code<end_of_turn>
<start_of_turn>model
Best code is no code.
Second best is code you can delete tomorrow.
I check this before every feature: can I solve this with a prompt instead of a codebase? Half the time, yes.<end_of_turn>
|
Write a LinkedIn post about: The cost of context switching | Context switching costs 20 minutes each time. You just don't feel it.
3 switches and your deep work day is gone.
I batch all messages to 2 windows. Mornings are for building. | <start_of_turn>user
Write a LinkedIn post about: The cost of context switching<end_of_turn>
<start_of_turn>model
Context switching costs 20 minutes each time. You just don't feel it.
3 switches and your deep work day is gone.
I batch all messages to 2 windows. Mornings are for building.<end_of_turn>
|
Write a LinkedIn post about: What agentic coding taught me about code maintenance | Agentic coding is fast. Maintenance is where it hurts.
The model is great at finishing the current task. It's not thinking about next month's feature.
So it picks the structure that works now, not the one that reuses nicely later.
We fixed this by giving it a simple rule: think about the next feature before you writ... | <start_of_turn>user
Write a LinkedIn post about: What agentic coding taught me about code maintenance<end_of_turn>
<start_of_turn>model
Agentic coding is fast. Maintenance is where it hurts.
The model is great at finishing the current task. It's not thinking about next month's feature.
So it picks the structure that ... |
Write a LinkedIn post about: Structure, naming, abstractions - my checklist | My checklist before any PR is short. Structure, naming, abstractions.
Structure: can the next feature reuse this without copy-paste? If not, structure is wrong.
Naming: would I understand this in 3 months without a comment? If I need a comment, I rename.
Abstractions: did we abstract after seeing the pattern twice? ... | <start_of_turn>user
Write a LinkedIn post about: Structure, naming, abstractions - my checklist<end_of_turn>
<start_of_turn>model
My checklist before any PR is short. Structure, naming, abstractions.
Structure: can the next feature reuse this without copy-paste? If not, structure is wrong.
Naming: would I understand ... |
Write a LinkedIn post about: What mastering AI-assisted coding actually feels like | Mastering AI coding doesn't feel like mastery.
It feels like you finally know what to not ask.
You stop asking for full features. You ask for small, verifiable steps.
You stop trusting output. You verify.
It's less prompting, more judgment. The tool did the same thing - you just got better at directing it. | <start_of_turn>user
Write a LinkedIn post about: What mastering AI-assisted coding actually feels like<end_of_turn>
<start_of_turn>model
Mastering AI coding doesn't feel like mastery.
It feels like you finally know what to not ask.
You stop asking for full features. You ask for small, verifiable steps.
You stop trus... |
Write a LinkedIn post about: One good example beats 10 instructions | One good example beats 10 instructions.
I stopped writing long prompts. Now I show one good output and say "do it like this".
Models learn from examples, not essays. | <start_of_turn>user
Write a LinkedIn post about: One good example beats 10 instructions<end_of_turn>
<start_of_turn>model
One good example beats 10 instructions.
I stopped writing long prompts. Now I show one good output and say "do it like this".
Models learn from examples, not essays.<end_of_turn>
|
Write a LinkedIn post about: What happens when you leave agentic coding unchecked | Left unchecked, agentic coding accumulates complexity. Quietly.
Each feature works. But together they create 4 ways to do the same thing, 3 naming styles, 2 error patterns.
No single feature is wrong. The codebase as a whole drifts.
We needed a rulebook not because the code was broken, but because it was becoming un... | <start_of_turn>user
Write a LinkedIn post about: What happens when you leave agentic coding unchecked<end_of_turn>
<start_of_turn>model
Left unchecked, agentic coding accumulates complexity. Quietly.
Each feature works. But together they create 4 ways to do the same thing, 3 naming styles, 2 error patterns.
No single... |
Write a LinkedIn post about: Explaining RAG to my non-tech friend | I explained RAG to my non-tech friend like this:
"You have a library and a smart friend. Your friend is smart but hasn't read your books.
RAG is giving your friend the right book page before you ask a question. Now he answers using your books, not just his memory."
She got it instantly.
If you need jargon to explai... | <start_of_turn>user
Write a LinkedIn post about: Explaining RAG to my non-tech friend<end_of_turn>
<start_of_turn>model
I explained RAG to my non-tech friend like this:
"You have a library and a smart friend. Your friend is smart but hasn't read your books.
RAG is giving your friend the right book page before you ask... |
Write a LinkedIn post about: Building a rulebook through iterations with Claude | I didn't write the rulebook alone. I built it with Claude.
Started with 20 rules. Tested on real tasks. Kept what worked, removed what didn't. 3 iterations.
Claude pointed out where my rules were vague. I made them specific.
Final version is 70% basic engineering, 30% lessons from our codebase.
Building with the mo... | <start_of_turn>user
Write a LinkedIn post about: Building a rulebook through iterations with Claude<end_of_turn>
<start_of_turn>model
I didn't write the rulebook alone. I built it with Claude.
Started with 20 rules. Tested on real tasks. Kept what worked, removed what didn't. 3 iterations.
Claude pointed out where my... |
End of preview. Expand in Data Studio
LinkedIn Natural 150 — gemma-3-1b-it style tuning
150 curated LinkedIn posts in a simple, conversational, twitter-like style (no cringe drama).
Composition: 44 twitter-gold short (<50w) + 84 core short-medium (50-80w) + 20 medium-long (80-150w) + 2 long gold. Avg 60.7w.
Format per line (JSONL):
{"prompt": "Write a LinkedIn post about: <topic>", "response": "<natural post>"}
Files:
data/final_dataset.jsonl— 150 full examples (use this)data/train.jsonl— 138 traindata/val.jsonl— 12 val (Gemma 3 chat formatted with )
Usage:
from datasets import load_dataset
ds = load_dataset("harsh-jos/linkedin-natural-150")
# or local line:
# ds = load_dataset("json", data_files="hf://datasets/harsh-jos/linkedin-natural-150/data/final_dataset.jsonl")
Training: QLoRA on google/gemma-3-1b-it via unsloth/gemma-3-1b-it-bnb-4bit, r=16, lr=2e-4, 2-3 epochs. See ml-exp/scripts/02_train_lora.py.
- Downloads last month
- 36
Models trained or fine-tuned on harsh-jos/linkedin-natural-150