Break a CSV export into checkable agent tasks
Your task
Your Java admin tool needs a CSV export. It queries the data, streams it to a file, uploads the file to S3 and emails the user a link to download it. A colleague plans to give an agent the whole feature in one prompt and merge whatever comes back.
How would you break the work up so you can trust each piece, and which piece would you not let the agent own?
It covers
- Decomposition for trust
- Checkpoints