For a 1-5 person business in 2026, AI consistently pays back on six task classes (drafting, summarising, scheduling, research synthesis, code generation for solo developers, image/asset production) and consistently fails on six others (high-stakes customer-facing decisions without disclosure, regulatory advice, complex multi-party negotiations, brand-distinctive creative work, anything requiring physical presence, anything requiring social proof of human authenticity). The 90-second test before delegating any new task: (a) if AI gets it wrong, what is the worst-case cost, (b) does the customer expect a human authored this, (c) is disclosure feasible without breaking the trust contract.

Operators register pillar piece. Anchor for the entire register — every operator-section piece can link 'see the delegation framework here.' Cadence 45-day (operators tooling cadence + framework refinement). Trigger conditions: significant capability shift in any of the six 'AI works' classes (e.g. multi-step reliability moving above 60%); regulatory clarification on disclosure obligations under EU AI Act Article 50; emerging consensus across operator-cohort surveys on the six 'AI fails' classes. Sister claims: OPS-024 (when not to use AI for small business — inverse), OPS-009 (picking first AI agent), OPS-018 (1-page AI policy).