Yes. Humans are also unreliable and nondeterministic (though certainly more reliable). Accordingly we have built software dev practices around this. I imagine it would be super useful for example to have a "TDD enforcer":

Phase 1: only test files may be altered, exactly one new test failure must appear.

Phase 2: only code files may be altered. The phase is cleared when the test now succeeds and no other tests fail.

If you get stuck, bail and ask for guidance