Depends on the quality of your validation loop. Can the agent find the bug in a five second unit test, or does it have to run the full deployment test?
It also presents tradeoffs in compute budget. Cycles spent executing large arrays of tests could mean less tokens spent debugging.