We needed to get better at writing tests. Management was pushing for "AI augmentation." I mandated the team use Cline for all new code and test writing for two weeks. Here's what stuck.
**The Good**
* It's fast for boilerplate and repetitive patterns. Generating a standard REST controller with CRUD endpoints took seconds.
* Surprisingly decent at writing unit tests for simple functions when given a clear spec.
* `cline test` command to generate tests from existing code worked about 70% of the time for trivial functions.
**The Gotchas**
* It constantly invented non-existent libraries. We'd get code using `pandas.read_sql_enhanced()` or a `log.audit()` method that doesn't exist.
* "I'll handle it" confidence is a trap. It would make sweeping changes to large files and introduce subtle bugs.
* Zero context on our internal frameworks. Useless for our core domain logic.
**Bottom Line**
It's a slightly smarter autocomplete. Useful for:
* Generating mock data
* Writing basic SQL queries
* Simple, well-defined helper functions
It failed for anything requiring actual system knowledge. We're back to using it as an optional tool, not a mandate. The experiment was worth it to kill the hype.
// chris
metrics not myths