I spend my days building things with AI. Websites, assistants, videos — the whole of Autonoma runs on top of it. So when a new model shows up, I don't read the announcement: I put it to work the next day, on real work, with real deadlines.
This month I worked side by side with two models from the same house — Fable and Opus — and the difference in the ability to simply get things finished was one of the most striking things I've seen all year. It's not that one is good and the other bad; both are excellent. It's that with one I describe the destination and it arrives; with the other I describe the destination and spend the afternoon giving directions along the way. For a one-person studio, that difference isn't comfort — it's payroll. It's the difference between delivering a client project in a night or in a week.
Not everything was smooth, and I think it's fair to tell that part too. There were days when everything went strangely slow. There were moments when I came back to an open conversation on one model and found another in its place, without having switched anything. And there was one curious day when the screen said one name and the agent, when I asked it directly, answered another — and the quality of that afternoon's work confirmed its answer, not the screen's. I was left not knowing what I was using, and my usage said one thing while the result said another.
I'm not writing this as a complaint. I'm writing it because I learned two things that hold for anyone who works with these tools — and, honestly, for any work at all:
First: trusting doesn't excuse verifying. I check my agents' work the way I check my own — with my eyes, on the real screen, on the real phone. When something comes back wrong, the cause is rarely ill will; it's almost always an invisible gear nobody looked at. Asking "what's actually going on here?" solved more problems for me this month than any better prompt.
Second: the right tool is worth what it costs. We're a small studio. Every subscription weighs. And even so, the month's conclusion was this: there is work that's only worth doing with the best model available, even when you pay for it. Because the real cost isn't what you pay for the tool — it's what you pay in hours, in rework, and in the energy that should have gone to the client.
There's an old principle I try to apply to everything Autonoma does: to point things out with the same care I'd want someone to use pointing out mine. To tell the truth without wounding, to thank without flattering, to correct without humiliating. The people building these tools are doing something genuinely hard and genuinely new — and my gratitude is real: this month I delivered things that a year ago would have taken a team. The odd details? I note them here the way you'd tell a friend the balcony light was left on. It's how we'd want it done to us.
If you run a small business and you're deciding whether "AI" is overkill for your size: it isn't. But choose the tool the way you choose a supplier — by what it gives you back in life, not by the price on the label. And always check with your own eyes.
— Alex