Writing
What it takes to trust an agent.
Told through the agents I run at home. Posts go out on Substack; every one is here too, searchable.
Episode 1 of "Where do we eat?" publishes this week. It will appear here the day it goes out. Subscribe to get it by email.
No posts match. Try another word, or browse the full archive on Substack.
The series
Where do we eat?
Every family asks it three times a day. I built an agent to answer it, took it on a family trip to New York and Boston before it was ready, asked every question twice, and logged the difference. Each episode opens on a meal and closes on one thing anyone running agents in production has to get right.
- This weekMy food agent wasn't ready for our trip. So I made the trip its test.
- ComingRecord once, replay forever
- Coming“Boring”: when every answer passes and none is worth the trip
- ComingThe model chooses. It is never the source of a fact.
- ComingThe judge has to earn its vote
- ComingThe user's choice is the label
- ComingFinale: seven questions an AI agent has to answer before anyone lets it act
Get each new post by email
The writing goes out on Substack. Every post also lands here, searchable.