8 ms·
> First, I use copilot to complete test cases and boilercode – I don’t trust it for logic Very interesting! I want to write the test cases myself. When copilot
by tasuki 2y ago
> First, I use copilot to complete test cases and boilercode – I don’t trust it for logic
Very interesting! I want to write the test cases myself. When copilot writes some code that passes my tests, I think I can trust it.
If you don't even trust copilot to write your logic, how can you trust it to write the specification for how the logic should behave?
- sevg 2y agoI think it's easier to review test cases as they tend to have less code, less complex logic, and less interaction with other code.
- tasuki 2y agoYes of course. But wrt trust, it's vital for the specification! If the tests are good enough, the implementation is an implementation detail!
- sevg 2y agoWell my point was that if you're starting off by writing tests, AI assistance is low-risk: the test code should be simple and concise enough to review very quickly. So you're not blindly trusting. Tests also tend to have a repetitive and clearly defined structure, an environment that AI thrives in. (Also, devs usually find writing tests boring, while AI never gets bored.) And once your tests are in place, if desired you can then get AI to write the actual code too.