Real systems running in production: AI agents, fintech infrastructure and automation. No tutorials; what I build and what I learn operating it.
Onda Corta publishes daily with nobody pressing buttons. Why quality control lives in code instead of the prompt, and what the human actually does.
I built a tool that measures whether ChatGPT, Gemini and Claude recommend a brand. Four measurement problems I hit with data, and what I did about each one.
Someone on X asked how I limit the agents that run unattended. The five layers I use: small tasks, code gates, output validators and manual publishing.
My nightly bug hunter runs at 2am over six repos, fixes what it can prove and opens a PR. The design, the numbers and the failures behind every guardrail.
How I designed a fleet of ~20 AI agents that monitors, diagnoses and repairs my production infrastructure, and why none of them can touch money.