On product building

A good technical tradeoff analysis comes down to whether your future self is willing to live with the solution. Say you are analyzing a problem and you have two solution options. Option A is technically superior to B. It's more novel, has new technical challenges that the team

DRY principle translates really well while working with AI agents. Beyond code reuse which can save you a lot of tokens, is "unit of work" reuse. The idea is simple. Write a clear text file that instructs ai agent to do a single unit of work and let

There are good and bad reasons to write a yearly roadmap. The bad reason is to plan your entire year. Those plans rarely survive reality. The good reason is to help you put everything into perspective, ask hard questions, and articulate what you want to do with your people, your

I've been noticing many products adding AI chat interfaces lately. But are they actually better than a well-designed UI? We humans can process visual information in roughly 13 milliseconds. Our average typing speed? About 40 words per minute. Yet many teams seem to use chat as a shortcut

Generations of engineers hated to write documentation. Now the dominant use case for AI coding agents is writing documentation. Documentation felt painful because it was backward. Engineers needed to build the software and then explain what they did (Build → Explain). Now it's forward: developers explain what they want,

The point of experimenting is to get it out of your system. If you are thinking of something and wonder whether it's going to work, try it, build it, work on it, write it; do whatever makes it real. Best case, it succeeds; worst case, it fails miserably,

I've been trying different methods to validate product ideas over the years. One lesson I learned is that many methods are ways to avoid launching the product. It's a scary step for many of us building products to find out if what we built is going

I've noticed indie hackers and startups are getting far more from AI than larger companies. Developers often point to codebase size - it's easier for an LLM to grasp 10K lines of code (LOCs) than millions. That's true to an extent, but it misses

One of the things I'm curious about is how AI will change the relationship between developers and their code. Most engineers love their code. If you want to put this to the test, just go and create a PR (pull request) on someone else's code and

Technical debt isn't a technical problem. It's everyone's problem. Yet there's this ongoing tension between engineers and the rest of the organization about balancing technical debt with feature delivery and speed. Most people still see it as purely an engineering concern. The