Python performance debugging is exactly the kind of concrete coding task AI handles well.

Performance debugging of Python scripts follows well-established patterns — profiling, identifying bottlenecks, applying vectorization or chunking. The structure is consistent across instances even if the specific script varies.

Success is measurable: the script should run faster, ideally with a target time stated or implied. The agent can benchmark before and after, making completion criteria concrete.

The agent needs the actual Python script and ideally a sample of the CSV to profile accurately. Without the script, the agent can only offer generic advice rather than targeted fixes — this is the key dependency.

The agent is proposing and explaining changes, not deploying to production. The human reviews and applies the fix, so mistakes are easily caught and reversed before any real damage occurs.

Performance optimization is largely algorithmic — profiling output, complexity analysis, and library selection don't require intuition or taste. A human should review the final changes, but the diagnostic and proposal work is well within AI capability.