AGENTIC AI
Three years of shipping agents into production.
Most teams are still building their first agent. I have spent the last three years learning what breaks at scale, what survives contact with real users, and what to do when the model is wrong at 3am.
3 years
Building agents before it was a category
Production
Running for thousands of ad accounts
Weekly
Ship cadence into customer environments
WHAT I'M THINKING ABOUT
Agent architecture — Orchestration patterns, MCP, when to build vs adopt.
Evals & reliability — How we test what we cannot fully predict.
Cost at scale — Keeping inference bills survivable as you grow.
Trust & control — Giving agents authority customers will accept.
SELECTED WRITING
The Weird Claude Code Error That Ate My Afternoon (and How I Fixed It)
Today, while I was deep in the middle of handing off some work between Claude Code sessions, I got this weird error...
Read →Building agentic AI in production?
I do a small amount of advisory work with engineering teams shipping their first agents. Always happy to compare notes.