My preferred approach in many projects is to do some unit testing, but not a ton, early on in the project and wait until the core APIs and concepts of a module have crystallized.
At that point I then test the API exhaustively with integrations tests.
In my experience, these integration tests are much more useful than unit tests, because they remain stable and useful even as you change the implementation around. They aren’t as tied to the current codebase, but rather express higher level invariants that survive refactors much more readily.
Recent articles
- The Summer of Johann: prompt injections as far as the eye can see - 15th August 2025
- Open weight LLMs exhibit inconsistent performance across providers - 15th August 2025
- LLM 0.27, the annotated release notes: GPT-5 and improved tool calling - 11th August 2025