Tests are a great use for AI coding, lately. Six months ago Claude Sonnet was writing tests that always passed without testing anything, that has improved dramatically with Opus 4.5/4.6 - it actually hits the functionality now, not just code coverage.