ai-fluency-delegationlisted

# Delegation — scoping authority before the work starts Delegation is deciding *whether, when, and how* to engage an agent — and making that decision **explicit before** the agent acts, not correcting it after something unexpected. It has three parts. ## The three components 1. **Problem awareness** — What is the actual goal (not just the immediate ticket)? What subtasks does it require? What are the stakes and how reversible is each decision? 2. **Platform awareness** — What does this agent do well here? Where is human judgment irreplaceable? What are the relevant limits (hallucination, scope drift, context length)? 3. **Task division** — Which subtasks run autonomously, which are collaborative, which the agent governs over many future interactions — and where are the human checkpoints? Drive autonomy by **stakes × reversibility**: low-stakes and reversible → more autonomy; high-stakes or irreversible → more checkpoints. ## The three modes - **Automation** — the agent executes a well-defined task; the human checks the output (format code, write a test, add a check under contract). - **Augmentation** — human and agent work back and forth (design decisions, tradeoffs, architecture). Judgment stays with the human; breadth comes from the agent. - **Agency** — the agent governs future interactions on the human's behalf (recurring workflows, long-running runs). Needs the most upfront Description and the most ongoing Discernment. ## Authority-scope declaration