GitHub Actions Isn't a CI
GitHub Actions runs your tests, so technically it's CI. It's still not a CI system, and the whole industry spent a decade mistaking one for the other.
Two decades of building software, shipping startups, and open source. Opinions included.
GitHub Actions runs your tests, so technically it's CI. It's still not a CI system, and the whole industry spent a decade mistaking one for the other.
AI didn't make me faster. It made me the bottleneck. Running five to twenty agents across every function of my company, the slowest part of the whole thing is the one part I can't optimize away: me.
For fifteen years, the smart move was to rent your primitives from GitHub. A decade of neglect and a wave of AI just flipped that math. You don't displace GitHub. You stop depending on it.
Mitchell Hashimoto says merge queues fall apart under AI agents. I run a merge-queue company, and he's half right. The queue isn't the thing that breaks.
AI agents review my pull requests now, not me. But there is one thing I will never be able to hand them: everything my company knows that was never written down.
A 95-year-old French furniture maker out-supports a French cloud company. The gap in European tech was never the technology. It's the service mindset.
A rewrite used to be career suicide. Porting our CLI to Rust took a month of review and a model that knew the language better than I ever will.
Armin Ronacher's 'Build It Yourself' got reposted after the npm attacks. The advice is right. The hard part is the question nobody quite answers: which dependencies do you actually use?
Frontier LLMs corrupt 25% of what you delegate. The fix isn't going back to writing by hand. It's the same linters and CI we built for humans, finally pointed at the new worker.
Free dinners from Foodora ended. Free Claude Code dinners are next.
Every code review tool I've used in twenty years has tried to answer the same question. AI just made the answer unavoidable.
The software pipeline has always had a bottleneck. For decades it was writing code. AI fixed that. Now it's the step right after.
Every team using Alembic with parallel branches hits the same wall. We built a library that uses git commit history to fix it.
I had a real business conversation over email. Turns out the other side was an AI agent. I kept talking.
Wall Street wiped $300 billion from SaaS stocks and declared the model dead. They're right about the wrong thing.
The framework hasn't changed. The weight of each skill has.
AI writes code 10x faster than humans. CI still runs at the same speed, fails for the same flaky reasons, and costs more every month. Something has to give.
I used to hold entire systems in my head. Now I hold seven terminals. The trade was worth it, but something real got lost.
Code generation got cheap. Review didn't. That asymmetry is destroying open source faster than any AI policy can fix.
The old open source deal is dead. What replaces it isn't a fix, it's a transformation. Open source stops being a community and becomes a supply chain.
AI broke the implicit deal that sustained open source for 30 years. Usage is up. Engagement is gone. The economics don't work anymore.
I dug into Entire's open source Checkpoints CLI. It's a clever abuse of git internals — shadow branches, orphan metadata, and a session state machine. Here's how it works.
The former GitHub CEO just raised $60M to rebuild developer tooling for the agentic era. He might be right that git needs a rethink — I've been hacking around the same problems.
I've been coding for 25 years. Since January, I haven't written a single line. And it feels like relief.
In a few years, the only proof something is real will be that it existed before AI did.
Everyone fears for the juniors. But the engineers who stopped growing at the wrong layer have more to lose.
The hard part isn't building. It's everything else.
GitHub Actions Pricing: The Platform Reality Check
Why I’m worried and why founders can’t afford to wait for Europe to wake up.
AI feels like 1999 all over again
Reflections on building, leading, learning, and staying sane along the way.
How I built Mergify’s new autoqueue in less than an hour a day
And I Might Have Killed It
How Spotify just confirmed the AI content tsunami I predicted.
What It Taught Me About Support
Why most teams would be better off with a clear plan than a quarterly ritual.
AI Isn’t Broken, Our Expectations Are
There’s more than one way to be a founder.
Quality is slow, hard, and totally worth it
Growth Is the Battlefield
Or Why AI Isn’t Google
The alarming state of security in too many tech companies
Stop Settling for the Crumbs
How CI Insights Was Born
How we hire at Mergify
The Product That Never Shipped
“It’s Complicated” Is Not an Excuse
The Moment We Realized CI Optimizer Was Doomed
And How to Fix Them
How We Spent 6 Months Building a CI Tool Nobody Asked For