We had an old Node.js service, no tests, nobody on the team wrote it. Tried writing tests manually for two weeks and gave up. Too many moving parts to mock by hand. Keploy attached to it with no code changes; we ran real traffic through for a couple of days, and came out with over 130 tests. Actually reflecting how the service behaves, not how we guessed it did. Review collected by and hosted on G2.com.

If real traffic never hits a code path, that path won't be in your tests. Makes sense, but we didn't realize we needed to deliberately trigger edge cases during recording. Would've saved us a second session. Review collected by and hosted on G2.com.