Signal / Noise
A working ledger · Product in the time of AI

I build AI products for the enterprise — agentic systems, generative AI, production ML — where the hard problems are rarely the models, and almost always trust, evaluation, and what an organization will actually adopt. This site is where I keep what I've learned, in the open, for anyone it might help.

Thoughts
The eval is the spec
The model is the easy part
How I decide an agent is good enough to ship
All thoughts →
How to make AI products

Over the past few years I've organized what I've learned shipping AI in production into a course — 24 units, free, no sign-up, no expectations. Start with Unit 01: What AI Engineering Is (and Isn't), or see the full syllabus.

What I'm working on

Everything I build lives in the open. The current work — datasets, experiments, half-formed ideas — is on GitHub.

My reading list
The Alignment Problem — Brian Christian. The clearest account of why "the model works" and "the model is right" are different claims.
Thinking in Bets — Annie Duke. Decision quality under uncertainty; quietly, the whole job.
Prediction Machines — Agrawal, Gans, Goldfarb. The economics of cheap prediction, still underrated.
High Output Management — Andy Grove. Most AI strategy is Grove with new nouns.
AI Engineering — Chip Huyen. The best current field guide to building with foundation models in production.
The full list →
Jobs

Product Careers is a system I built because job boards are a noise problem: roles aggregated from career pages and the major boards, deduplicated, categorized, stale postings aged out. If you're a PM looking, it's yours — read how it works.

Carlos Rivero Say hello on LinkedIn →