For months I built my business by treating ChatGPT as a coach.
I had no real idea what I was doing. I'd played with n8n, the automation tool everything now runs on, a couple of times, but I'd never designed a workflow from scratch, never started from an architecture and built out. The whole thing was new. So I just asked. How do I rebuild this automation in n8n? What goes wrong if I do it this way? And we talked it through, for weeks, then months.
I learned a huge amount like that. I still think large language models are the best learning tool I've ever had, mostly because they cover the gaps in what you know and let you vent the frustration and find the fix in the same breath. That part I won't take back.
But at some point I started going in circles. The same loops, the same half-fixes, the same problem coming back wearing a different hat. I'd solve something, it would un-solve itself, we'd go round again. I tried Gemini. Still circling.
I ended up at Claude mostly out of frustration, and the work got easier almost immediately. The connectors, the skills, the way it held a thread across a long piece of work without dropping it. I've been an Anthropic believer since.
The moment it stopped being a toy
The real shift came with Opus. Before that, everything was short feedback loops. I prompt, it answers, I correct, it answers, round and round, with me babysitting every step.
Then I gave it a single thread with one clear prompt about what I actually wanted, and it just kept working. Independently. It didn't stop to check in every two minutes. It carried the task.
That was the moment it stopped feeling like a cool toy and started feeling like something that changes how a person can work. I was building Memolio entirely on my own, a personalised illustrated book for grandparents, and suddenly the thing helping me build it could hold a real piece of work without me hovering over it. I remember thinking, clearly, this is going to change the world, not just be a clever thing to try out.
What I'd tell someone still stuck
I'm not here to tell you which model to use. Tribal loyalty to a tool is a waste of energy, and the models keep leapfrogging each other anyway. What I'll say is narrower and more useful: if you've been stuck in the same circles for weeks, the tool might genuinely be the variable. It's worth testing. Mine was.
The deeper point is that none of this works without you staying in the loop. The model can carry a task. It cannot decide whether the task was worth doing, or whether the output is actually right. That judgment is still yours. It's the whole reason I, one person with no engineering background, could build a real business out of subscriptions and stubbornness.