So today, I needed to write unit tests for a new API endpoint on the Lightswap server. Instead of writing just a few use cases i decided to review all of the unit tests and I have and basic review them and revamp them (no small task).

The Challenge We need a thorough suite of tests testing all endpoints and more, non-functional requirements, think speed, handling errors, etc

Doing this manually would easily take 1-1.5 days of focused development time.

The Benefits Speed: What would take a full day of manual coding was completed in about 45 minutes of collaboration with Claude Code.

Comprehensive Coverage: Claude helped identify test scenarios I might have missed, like Unicode handling, rate limiting behavior, and edge cases with special characters.

Consistency: All tests follow the same patterns, naming conventions, and structure without the inconsistencies that creep in during manual development.

Documentation: Claude generated strong documentation explaining how to run the tests, interpret results, and troubleshoot issues.

A human review is needed. This is fine but it means my focus is more on finding holes rather than determining the coverage or typing the unit test.

The AI has access to the code but that does not mean it understands the context or nuances, or gotchas that I'm aware of with the code.

Optimizing for the wrong thing I realised it was initially optimizing for completeness/coverage and cost whereas I'm optimizing for correctness and manageability (50 great tests are better than 150 ok ones).

The cost of running tests that are making API calls to Claude.

Domain Knowledge: The AI will need guidance on bitcoin-specific scenarios and wallet application workflows.

The outcome? Claude wrote the code in minutes whereas it would have taken me the whole day. Tomorrow and Monday I'll review the tests and report back on the results and see how good it was!