The work
Tools for building carefully.
Experiments in making automated work understandable, accountable, and useful to the people around it.
Different jobs. One standard.
These are not claims that autonomous systems should be left alone. They are attempts to give new kinds of work a legible history, a place for review, and a person who remains accountable.
Make the work replayable.
Flowbee is an orchestrator for fleets of AI coding agents. It keeps the control plane deterministic and replayable so automated engineering can remain understandable.
Let useful memory persist.
Pearl is a memory layer and model router: an experiment in helping agent work carry forward what it has learned instead of beginning from zero every time.
Keep a review loop.
Polly supervises agent teams. The point is not a machine that gets the final word. It is a clearer way to see, question, and improve what a team of agents is doing.