What I like and worry about with AI

I use AI daily. The hype is exhausting at this point. Product demos feel desperate. A lot of features labeled AI are just god-tier autocomplete with better marketing.

The tools still help. I can test five different approaches in ten minutes. I can get answers immediately. I can move through a codebase in a couple minutes.

It removes friction from repetitive work. I do not need to write the hundredth config file or boilerplate test by hand. The boring parts go away, which leaves more energy for the parts that need thought.

The ethics are still bad. Training data gets scraped without consent. Models burn through absurd amounts of resources. Ugh.

Codebases rot when people accept suggestions blindly. Code gets committed that nobody reviewed. Six months later, the team is debugging a brittle mess or the thing never even made it to production.

I have one simple rule. The output needs to be understood and verified. Every line gets read. Most of it gets rewritten. AI is like an intern who needs supervision and constant guidance.

If you cannot evaluate the output, you are in trouble. LLMs fail all the time. Debugging is impossible when nobody understood the original logic, and quality gets worse with every change.

AI works well for sketching ideas, moving fast, and getting unstuck. I would keep it there.