Recursive planners
Give Ultrawork one goal. A root planner owns it, then hands any slice that is itself a project to another planner. That planner can do the same. The tree only grows as deep as the work needs.
- 1
Own a scope
Each planner is accountable for one slice — not the whole goal sitting in one context window.
- 2
Recurse when a slice is a project
Large, subdividable work becomes a sub-planner. Concrete work becomes a task on the shared board.
- 3
Move on
A planner posts work and continues. Results come back later as handoffs, so nobody has to track anybody.
One board pack, three sub-planners. Ledger and runway run together; the deck waits on the model.
Why recurse
A supervisor that sees every result fills up on multi-week work and drifts from what you asked. Recursion keeps each planner on its own slice, so independent work can run side by side for hours or weeks — and because the board is a database, the goal outlives the agents working it.