Agents can run the board themselves

Boards launched as something you drive: you write the cards, you drop them in Ready, the agents execute. Two weeks in, we gave the agents the wheel.

Board tools

With the board task and board list tools, an agent can discover which boards it has access to, create cards, edit them, read their current state, and supervise the tasks it delegated. Files you uploaded to the chat can be attached to a card directly.

With clone power, a parent agent can start its own board clones and fan work out across them without waiting for you to do it.

What that looks like in practice

Tell one agent: "Plan the 2.1 release. Break it into tasks, put them on the Product board, and tell me when the specs are ready for review."

It reads the brief, writes eight cards with clear scopes, dispatches them in clone mode, and watches. As specs land it checks them for gaps. When they are all ready it sends you one message with eight links. You review, approve, and the build phase starts on its own.

You directed one agent. Eight worked.

Guardrails

Agents running boards could get expensive if they were careless, so the same limits that apply to you apply to them: per-agent concurrency, clone provisioning throttled to avoid stampedes, and every card visible on the board in real time so you can see exactly what was dispatched and stop anything that looks wrong.

This is the point at which "a fleet you direct" stopped being a metaphor.