Maestro

The problem

A studio that runs on AI agents has two failure modes nobody warns you about. Long-running background work is invisible: it either finishes or quietly fails, and you find out which one later. And every new AI session starts from zero: it has to be reminded where the files are, re-told the same context, pointed at the same documents, every single time. We were spending real hours re-introducing our own business to our own tools.

What we built

Maestro is the answer to both. Every automated or scheduled task records itself as a run and reports progress in real time: step-based progress bars, elapsed time, an ETA, streamed live. A run opens when work starts, ticks forward step by step, and closes as done or errored, so nothing slips silently. Around the runs sits the organized layer a session needs: the documents, the project state, the operating rules, all in one place any tool can read.

What changed

It now tracks everything we do: on the order of 100 runs across four apps shipped or heading to the App Store, every business we operate, and all of the consulting. The re-telling stopped. A session finds the right context on the first try because the context has one home. This is the system behind the line on our homepage: we run our own companies on what we build, and this is the thing they run on.

It is also the honest demo of the ops audit. What we map in a client's business is what Maestro already does for ours.