Claude Opus 4.8 launched yesterday. It's good. Try it out.
But lately, in our workshops, a lot of you have told me: you're tired of the AI model race.
Nobody wants to wake up every week and ask: Is Claude still best? Is Codex better now? Should we be testing Cursor? Are the Chinese models good enough yet at a tenth of the price?
We all just want a simple answer. Pick the tool, train the team, build the workflows, get back to our lives.
That instinct makes sense. But there's a bigger risk here than picking the wrong tool.
AI is still in its subsidy phase. The price you pay isn't the real price. It's the cheap-Uber era, before the rides cost what rides actually cost. And that era is ending.
You can already see it. GitHub has moved to usage-based pricing. OpenAI has signaled that unlimited plans probably won't last. The real cost of running frontier models keeps creeping up. None of these is dramatic on its own, but the direction is pretty clear.
So your AI costs could climb quickly, and the workflows you lean on could get expensive overnight.
My advice is the boring kind: don't commit everything to one tool. Diversify, stay curious, and keep testing while the market is still moving.
Here's what that's looked like for me lately. Six weeks ago I moved most of my work from Claude Code to OpenAI's Codex. Ten days ago I picked up Cursor's Composer 2.5, which is blazing fast. And today I'm back in the Claude app part time, with the new Opus 4.8.
So until the market is stable, stay nimble and don't rely on a single tool.
What's in your AI stack?