Four new AI models shipped this week. I switched none.

Four new AI models shipped this week. Anthropic, OpenAI and xAI all put out something faster or cheaper in the same few days. One year ago that news would have had me up all night switching all my work to the newest one. This week I switched nothing, and here is why.

Why not just upgrade?

The newest Claude model is genuinely good, and cheaper for an intro period. But the upgrade also breaks a few things. Some settings my code used to rely on now return errors, and the same text costs about a third more to run. Swap it in across everything without looking and you find that out in production, at the worst possible time.

So the move is not to chase the release. Test the new model on one small job, see if it actually performs, and only then let it near the rest.

Which model should you even use?

So many people ask me this. Should I use this model or that one, ChatGPT or Claude or Perplexity or something else?

To be fair, the honest answer is that it mostly does not matter. The brain you use is important, of course. But whether you build on Codex or Claude, I would not say it makes a huge difference. You can make it work with both.

So what does matter?

Your infrastructure. How you have actually built out your systems. A different agent checking every build. A human gate on anything that spends money or goes public. A way to swap one part without the whole thing wobbling. That is the part that decides whether AI runs your business or just impresses you for an afternoon.

When a new model drops, that scaffolding holds, and I get to test calmly instead of scrambling. Better models are great, and I adopt the ones that earn their place. But the system is the moat, not the model of the week.

I build AI agents that run my entire online business. Check them out at pelleforsman.com.