On the Industry

LLMs Are Useful

All LLMs are useful. They compress hours of library discovery and learning into minutes. Delegating that research is a massive productivity gain.

On Production

LLMs are still not profitable. Smaller language models are way more profitable because you can measure their results. If you have the budget, use LLMs in production. But if your service actively depends on their usage, I cannot guarantee your success.

How I Work

I delegate research and boilerplate to LLMs, but I always apply human judgment. When Mimo tried building a custom static routing solution, I had to push back and ask it to check if Preact already supported that — and it did, with pre-rendering.

My Workflow

AGENTS.md

Every project gets an AGENTS.md — or several. Each project is different, each one requires its own configuration. My goal is to set it up as if a human developer were going to read it and follow it.

Code Quality

Linting, formatters, every available tool for code quality. The agents loop around that criteria. The guardrails are the project's own standards, not the LLM's defaults.

100% Review

I review all code produced. Git diffs are my filter. Context is kept layered — agents work within their scope, I keep the big picture.