Decompose for verifiable trust
You need to add a CSV export feature to an internal admin tool in Java: query the data, stream it to a file, upload to S3, and email the user a signed link. A colleague plans to give an AI agent the whole thing as one prompt and merge whatever it returns. Propose how you'd instead decompose this for the agent so that trust is verifiable at each step — and identify which single sub-task you'd refuse to let it fully own.
review_export_plan(step_lines: list[str]) → list[str][["query_and_row_mapping|1|0|60","csv_streaming|1|0|90","s3_upload|1|0|40","signed_link_email|1|1|30"]]out["query_and_row_mapping=delegate","csv_streaming=delegate","s3_upload=delegate","signed_link_email=human_owns"][["whole_feature_one_prompt|0|1|300","row_mapping|1|0|60"]]out["whole_feature_one_prompt=human_owns","row_mapping=delegate"]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.