Hire for the job, not forever. A worker exists for one task and costs nothing once the work lands.
a one-off migration
a flaky-test hunt
a benchmark run
Hire for the job, not forever. A manager starts a worker for one task, and when the work lands the worker reports back and goes quiet. A quiet worker runs no turn, so it calls no model and costs nothing, yet it stays on the team with everything it learned, ready for the next job.
The chat says ◆ started @flaky · its tab is on the strip, and the new worker's conversation opens behind the one you are in. The worker reports finished when the fix lands.
P shows the finished worker idle on the team, and one directive with Enter gives it the next job. Keys: PEscSend @notes a directive: use your ask tool to ask whether prices should round half up or half to even, then wait.Enter
If you set the manager's approvals to ◇ asks with Alt+A, it asks you first on a card, ◆ manager wants to start @flaky.
A worker is an ordinary conversation with a handle on the team. It spends while it works. When its turn ends it tells the manager in one Traffic line, finished, stopped or failed, and that line wakes the manager. From then on the worker is idle: the members card (p on the teams page) lists it as idle, and it spends nothing until someone gives it work.
The manager can give an idle worker its next job with a directive, which starts a new turn in the same conversation, so the worker keeps its context. Or it can leave it be. Nothing retires or deletes a worker on its own, and stopping one only ends its current turn.
Any manager can hire, at any level of the chart. A worker started by a sub-team's manager reports to that manager, and its result climbs from there: in one run a worker two levels down finished, its lead woke and posted the number up, and the team manager woke and told us, with no message in between.
Wrapping up
When the job is the whole team's, ask the manager to wrap up. It tells every member to finish the piece in hand and commit, then brings you a closing report: what was done, what is left, where the files are and what it spent.
Things to build
A research pod: a manager starts one worker per open question, and each goes idle with its answer, costing nothing until the next question.
A reusable specialist: keep the worker that wrote the test suite, and send it the next suite with a directive, context intact.
A counter two levels down: a sub-team lead hires a one-off worker to count or check something, and the answer reaches you without a message from you.
A per-change review pair: a manager hires a builder and a reviewer for a change and merges only on the reviewer's pass. Review every change