Notes on becoming an AI engineer the hard way
This is a journal of a slightly mad project: teaching myself to build properly with AI, the slow hands-on way, and writing down every bit of it as I go – the small triumphs, the things I get embarrassingly wrong, the afternoons that vanish into a problem I don't understand. You know those people who buy a falling-down house in the middle of nowhere and decide, with no real qualifications, to restore it themselves? That energy. Welcome to the building site.
I'm not doing this alone. I have an AI tutor – Claude, specifically – and I lean on it constantly: it explains things, it asks me questions back, it stops me before I do something daft. What it doesn't do is make the calls. I choose what to build, I decide what "right" looks like, I'm the one who says "no, that's wrong." I'm not going to pretend I arrived knowing any of this. That's rather the whole point.
I'm not a builder by training. I'm a marketer – but I've been tech-y by nature since long before any of this: the kid who took the radio apart to see how it worked, then put it back together with a few entirely mysterious screws left over. And I work at a busy little place where, if you're paying attention, you can see all the cracks – the processes held together with tape and good intentions, the jobs that swallow a whole afternoon for no good reason. Nobody hands you a tidy brief; you just watch the machine and notice where it grinds.
So I started fixing the worst bits myself – building small tools where there'd only been manual work and guesswork. A scheduling thing here, a tracking system there, automations to kill the copy-paste jobs that ate whole afternoons. One led to the next, and somewhere along the line I'd quietly built most of the systems that keep a busy operation running day to day – without ever once calling myself a builder. So I'm not starting from nothing. I've been doing this for a couple of years now, by feel and luck. What's new is that I'm finally learning to do it properly – because I'm not in this for the patching anymore. I want to become an AI engineer. That's the actual goal, written down in public so I can't quietly talk myself out of it.
I'm not doing it the conventional way, though – no quitting to go back to university for a degree. Quitting isn't the move: there's a job that pays the bills and a family that comes first, and the whole point is to make this change without upending a life that already works. Mostly, though, it's that I learn by building, and always have. So the plan is to make real things, break them, and write down what happened – with a few proper online courses lined up alongside to shore up the parts that building alone won't teach me.
The restoration channels I'm borrowing the spirit from all have one thing in common: they show you the mess – the collapsed ceiling, the load-bearing wall they knocked through anyway, the bit where they clearly have no idea what they're doing. That's the appeal. Tech almost never does this. You get the polished result and the clean demo, and none of the 11pm bafflement that came before it. So the bafflement is exactly what I'm writing down – the version with the mess left in. If you're teaching yourself something you've got no real business attempting yet, you're in good company.