AI Still Needs Supervision

I feel like the “old guy yells at cloud” meme.

I’m a heavy AI user, and not just for coding. The more I use it, the more supervision I find it needs. Models get more capable, but I keep catching the same mistakes.

The inconsistency makes it worse. One day, something amazing in one shot. The next, a stream of agreement and “you are completely right.” How do I know which day I’m getting?

Meanwhile, I keep seeing people say they don’t even look at the code LLMs produce. They just ship. I can’t remember the last time I didn’t ask for a change to something that looked fine now but would break a day or a week later: scale, memory consumption, parallelism, or another problem that isn’t immediately obvious.

It feels like working with a very intelligent, very insecure junior professional.

Everyone keeps saying this is done. I keep yelling.