Directing @yagni · Lesson 2 of 4
How does a Team's Playbook learn your rules?
Published June 9, 2026 · Updated August 14, 2026
The Playbook is the editable, team-shared set of rules a Team has learned for how its work gets done. Rules are captured from real work, your edits, approvals, declines, and corrections at the terminal, never configured up front. Responsibilities are what a Team is on the hook for; the Playbook is how.
What is a Playbook rule?
A plain sentence about how the work gets done. “Migrations ship expand-and-contract, in separate deploys.” “Webhooks retry through the queue, never cron.” “Anything touching billing gets the database lens on review.” Each rule is readable, dated, and traceable to where it came from, and together with Responsibilities it defines the Team: the mandate you write, the method that accumulates.
How does a rule get learned?
By watching what your team actually does, not by being configured.
| You do this | The Team learns this |
|---|---|
| Edit the same thing in two drafts | The shape that arrives right the third time |
| Approve a pattern repeatedly | The pattern is safe to handle routinely |
| Decline a proposal | The boundary it just found |
| Correct the agent at the terminal | A rule every developer’s agents pick up |
That last row is the harness working as a team: when one developer corrects an agent in YAGNI Code, the correction lands in the shared method, not in one person’s local setup. Your best reviewer’s judgment becomes the floor for everyone.
Who owns the rules?
The Team, on behalf of the workspace. A rule reads as the Team’s, “the Checkout Team learned to gate schema changes on a migration plan”, never as one person’s setting. When a rule lives in one person’s config, quality depends on who is on duty; when the Team owns it, new teammate, same Team, same method. The moat is not that the agent works; it is that your team’s judgment compounds in a place you can read.
What is a Play?
A bounded run of work the Team proposes from its Playbook: the goal, the steps, what it may spend, and the deadline, approved once and executed over days. Rules are the method; Plays are the instances. Approving a Play confers bounded authority for that run, every step still leaves a Receipt, and an ad-hoc Play that proves out becomes a new rule. The outcome publishes as what happened to the business, with the Play attached as evidence.
How does autonomy grow out of the Playbook?
Through the Ladder: Training, Supervised, Autonomous. In Training you review everything. Supervised means reversible work ships with an undo window while you watch the record instead of the queue. Autonomous means you review outcomes. Each level is earned on this Team’s record, per capability, and confirmed by a person; a revert drops the rung, and the re-earn goes on the record too. The floor never moves: irreversible and high-blast actions wait for a person at every level, with the audit trail to prove it (see security).
Common questions
Do I have to write the Playbook myself?
No. Rules are learned as a byproduct of the team working: edit a draft, approve a proposal, decline one, correct the agent mid-run, and the Team captures the pattern. You can edit any rule it learned, but the writing is not your job.
How is the Playbook different from the decision ledger?
The ledger records decisions: what was chosen, what was rejected, and why, and agents read it before they plan. The Playbook holds standing rules of method: how this Team does its work. A ledger entry is history that informs; a Playbook rule is an instruction that binds.
Can I see and change what a Team has learned?
Yes. The Playbook is plain, readable, and editable. Every rule shows when it was learned and from what, and an edit takes effect immediately: the Team works the new way from then on.
How does a Team earn more autonomy?
On the Ladder: Training, Supervised, Autonomous. Levels are earned per engagement from the track record and confirmed by a person; promotion is a button you click when the record supports it, never a switch that flips itself. Irreversible acts stay gated at every level.