Manual testing is focused human exploration of running software. Its distinctive value is pattern recognition: a person can notice that an output violates an expectation they never explicitly encoded as a test. section 8.3.1
A malformed heading reveals an unknown edge case
A markup converter may correctly turn # Header into <h1>Header</h1> and ## Header into <h2>Header</h2>. Trying #### exposes <h3># Header</h3>, prompting investigation of a failure that predefined tests might not include. section 8.3.1
Use manual testing as reconnaissance before committing to automated verification:
- Spend roughly 10–15 minutes exploring each significant piece of AI-generated code. section 8.3.1
- Start with the happy path to establish whether the basic behavior works. section 8.3.1
- Try unusual inputs, inspect the output, and follow unexpected results until the behavior becomes understandable. section 8.3.1
- Use the resulting mental model to choose automated tests that cover meaningful edge cases and failure modes. section 8.3.1
Manual testing is neither a complete verification strategy nor disposable overhead. It does not scale like automated tests, but it helps reveal what the tests should specify. Tests then express behavioral contracts rather than merely confirming implementation details. section 8.3.1 section 8.3.7
UI work is especially vulnerable when coverage is weak: human involvement may be needed during migration, while generating controller and UI tests reduces continued reliance on manual checking. section 2.7




