Greenfield versus state-machine tasks
Two tasks land on your desk. Task A: 'Build a small CLI tool with arg parsing, three subcommands, tests, and a README' — greenfield, clear shape. Task B: 'Touch the billing-state machine to add a paused state, threading it through 9 call sites with subtle ordering rules.' A colleague wants to fire off an autonomous agent for both and walk away. For which task is 'launch and walk away' appropriate, and how should your supervision differ?
choose_agent_leash(touched_call_sites: int, greenfield: bool, end_to_end_verifiable: bool, money_impacting: bool) → str[0,true,true,false]out"launch_and_walk_away"[9,false,false,true]out"review_each_checkpoint"Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.