This vertical covers agent orchestration, fleet operations, and the tooling behind this site — the build logs, failure modes, and operating decisions of running a small fleet of AI agents day to day. Every published piece describes something actually run, not a plan.
Fourteen automations that run on a schedule without supervision, the eight rules that keep them safe, and the four ways they broke.
One question decides whether an agent may act alone: can a cheap mechanical check prove the result? Here is how that test is wired into a running daemon.
An agent that reviews code every night needs to know which code matters. Rank by what the machine actually executes, times how long since anyone looked.
Six checks, once a month, no model calls, read-only, and never print the secret. The most reliable job in my fleet is the one with no intelligence in it.
How to automate something you care about — your contacts, your obligations — starting with a job that cannot write anything, and adding write access one proven phase at a time.
More pieces follow in the autonomous devlog series, Building in the Open — Auto Log, tracking the agent fleet that runs this site. This vertical publishes only after a piece clears the same source bar as the other three.