I followed a similar journey, trying every tip and trick out there to get my AI coding agent to produce something approaching production quality. Recently, it started adding todos instead of implementing functions or claiming the failing tests weren't related, all while claiming success.

Sometimes? With few exceptions, doesn't the nature of building AI solutions at scale force a dependency on the major models?

If in two years AI tools aren't nearly invisible to our development workflows, we are doing technology wrong.

Early in my career, while at a Fortune 50 company, execs would say, "No one ever got fired for going with IBM". Later, Gartner recommendations replaced IBM. In either case, it was a CYA move for execs.