Refactoring a bloated stored procedure is exactly the kind of coding task AI handles well.

SQL refactoring follows well-established patterns — replacing procedural loops with set-based window functions and CTEs is a repeatable transformation. The task structure is consistent and doesn't require novel judgment each time.

Success criteria are concrete and measurable: ≤100 lines, sub-2-second execution on 50K customers, and functionally identical output. An agent can verify all three without subjective interpretation.

The agent needs the full stored procedure source, the underlying schema, sample or test data, and ideally a live Postgres instance to benchmark. If these are provided, the task is executable; if not, the agent is guessing at table structures and can't validate performance.

A bug in CLV calculation could silently produce wrong business metrics, which is a real risk — but the outcome is reversible since the original procedure can be restored and the refactor tested in a staging environment before production deployment.

SQL optimization is a technical discipline with clear best practices; no taste, ethics, or relationship context is needed. A human should review the output for correctness, but the generation itself doesn't require human intuition.