AI boundaries in payments service
Your team owns a payments service in Go: an HTTP API layer, a core ledger/double-entry module, a fraud-scoring service, a pile of glue (CSV importers, internal admin tooling), and the test suite. Engineering leadership wants a one-page policy declaring which parts of THIS codebase AI agents may write freely, which require a human author with AI as assistant, and which are off-limits to agent generation entirely. Draw those lines for these five areas and defend the boundary you'd be most criticized for.
assign_ai_authoring_policy(blast_radius: list[int], trust_boundary: list[bool], machine_checkable: list[bool]) → list[str][[4,5,3,1,1],[true,false,false,false,false],[true,true,true,true,false]]out["human_authored_ai_assist","agent_draft_gated_by_checks","agent_draft_gated_by_checks","agent_free_rein","agent_free_rein"][[5],[false],[false]]out["human_authored_ai_assist"][[2],[true],[true]]out["human_authored_ai_assist"]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.