Playbooks

Playbooks are end-to-end workflows that turn the cognitive-load research into something you actually run. Each one protects the operator’s attention while agents work: a named routine, the research behind it, and the mechanics.

This section seeds with one; more follow the same shape as new patterns prove out.

---

Protect deep work while orchestrating

The problem it solves. Orchestrating agents is interrupt-driven by nature — an agent finishes, hits a question, needs a decision — and the naive setup services each interruption the moment it arrives. That shreds focus blocks (Csikszentmihalyi) and imports attention residue into whatever you switch to (Leroy). The result is a full day of shallow attention and degraded judgment on every thread.

The idea. Separate when work arrives from when you attend to it. Let agents run and accumulate their questions; batch those interruptions into a queue; drain the queue at natural task boundaries instead of mid-task. You stay in a focus block; the decisions wait for a stopping point.

1. Enter a focus block. Put the workspace into deep-work mode (aios mode deep-work). This signals that interruptions should be captured rather than surfaced immediately.

2. Let agents run and queue their asks. As agents hit questions or need approvals, they land in an asks queue instead of interrupting you. The queue holds the decision and its context, so nothing is lost — the closure Leroy’s research says you need before switching.

3. Reach a stopping point. Work your own thread to a natural boundary — a finished unit, a passing test, a committed change. Do not switch mid-task; that is where residue is worst.

4. Drain the queue at the boundary. Now, and only now, process the batched asks together. Batching similar decisions keeps you in one mode instead of context-switching per interruption.

5. Re-enter the focus block. Return to deep work. Repeat: run → queue → drain at boundaries.

Why it works, by the research:

• Batching-and-draining gives each task cognitive closure before you switch — Leroy’s direct remedy for attention residue.
• Protecting contiguous blocks preserves the conditions for flow — Csikszentmihalyi’s precondition for high-quality work.
• Draining at your boundary rather than the agents’ schedule keeps monitoring from becoming the continuous-vigilance task humans fail at — Bainbridge’s warning, answered with checkpoints instead of constant watching.

Note

The deep-work mode and asks-queue mechanics referenced here are part of the AIOS workspace toolkit’s operator layer. This playbook describes how to compose them into an attention-protecting routine; the cognitive-load lens covers the metrics that tell you whether it’s working (rising focus_block_avg_min, falling interrupts_per_hour).

---

More playbooks will land here — multi-agent review without vigilance decrement, setting leash length by risk, recovering context after an interruption — as each is validated against the research and in practice.