How to Know Your AI Actually Works
A practical guide to evals for small teams shipping AI features without a QA process.
Writing about building things with AI, product strategy, and pushing tools further than you think they go.
A practical guide to evals for small teams shipping AI features without a QA process.
What are the implications when your AI infrastructure is controlled by a foreign power? A US export order took a live model offline overnight, and Europe keeps treating sovereignty as a procurement problem.
A plain-language explanation of the architecture behind every major LLM, from attention to training to the questions nobody can answer yet.
I rebuilt my marketing agent from Anthropic's Managed Agents API into a visual n8n workflow. Here's how to build one that senses your traffic, checks whether AI assistants recommend your product, and hands you three prioritised actions every week.
The startup system is built to produce one unicorn. AI makes the case for the opposite: ten thousand small businesses run by people who already understand the problem, finally with the advisor they never had.
Every product instinct says remove friction. But as AI makes products invisible, the interesting design problem flips: where do you deliberately add it, and why?
AI compressed building from months to hours, but measuring whether anyone wants what you built still takes weeks. The Lean Startup loop is lopsided now, and nobody's found the new rhythm.
I ran 20 identical tests changing only one variable — task vs goal framing. The results reveal what actually makes AI behave like an agent, and it's not architecture.
On super ICs, Brooks's law, and where one person and AI beats a team.
I spent a weekend testing local AI models to see if I could work without cloud APIs. Here's what works, what doesn't, and what it costs to go off-grid.
Why some agent-native tools are homeruns, why others barely work, and what it means if you're building one.
Forget productivity. The best way to start with AI is to point it at something you actually enjoy.
I built an AI pipeline that pulls from 100 sources every morning, runs everything through Claude, and posts a curated brief to Slack. The whole thing costs $1.50 a month.
The market for simple desktop tools has always been built on an 80/20 compromise. That compromise is getting worse every month.
The indie founder ecosystem is full of tools that promise to show you what to build next. Most of them are selling the path, not the destination.
Enterprise champions aren't cultivated through account management. They're found — by building a product that matches how they already think.
What I'm learning by teaching AI to make videos.
Why good execution is no longer enough.
Delegation, AI, and the blank page problem.
Scaling without losing what made you special.
What happens when your vision gets passed from person to person.
You've forgotten what it's like to be new.
The difference between understanding a framework and owning it.
Stop debating whether ideas will work. Ask what would have to be true for them to succeed.
How startups deliver enterprise contracts without becoming consultancies.
Why startups shouldn't worry about technical debt, feature parity, or balance.
How to know what roadmap flexibility you actually have.
Lessons from building for users you'll never meet.
Product discipline from a century-old Cadillac.
Great products win by handling the edge cases others ignore.
How to use conference conversations to quickly test language, customers, and product bets. And bring back real insight.
Big contracts come with big compromises. You can still avoid roadmap takeover if you plan it right.