Notes from buildingthe operations layer.
Writing about running AI coding agents in a team: what breaks, what we chose, and why. Some of it is product thinking and some of it is engineering, and we label which is which.
// No growth-hacked listicles. If we have nothing to say, we do not post.
Everything we have written
Filter by topic, or read straight down. Newest first.
Why unattended agents stall, and what it actually costs
A permission request takes two seconds to answer and can idle a run for six hours. The gap between those numbers is not a technical problem.
Memory that survives a tool switch
Teams do not standardise on one coding agent, and they change their minds every few months. That is an argument for putting memory behind a protocol.
An agent took an action. Who authorised it?
Supervised autonomy only means something if the supervision is attributable afterwards. What that requires from a control plane.
Other things worth reading.
Changelog
What shipped, when, tagged by product. Release notes rather than announcements.
Guides
Step-by-step walkthroughs for setting things up and getting them working.
Glossary
Definitions for the vocabulary this category invented rather quickly.
Reading is free. So is deciding later.
When you are ready, an organization, a subscription, and one seat is the whole setup.