Most of what you hear about AI is framed as work. Prompting, automating, optimising. How to be more productive, how to replace workflows, how to do more with less. If that framing doesn’t excite you, or worse, if you’ve tried it and walked away frustrated, I don’t think the problem is you. I think the frame is wrong.
The best way I’ve found to use these tools is to point them at something I actually enjoy and see how far they go. Not something useful, not something productive. Something fun.
A friend of mine recently told me he’s worried that every job is “turning into a prompt.” He sees people around him getting excited about AI, building things, moving faster, and he doesn’t feel the pull. When he tries, he can’t get it to do what he wants, and the gap between what everyone else seems to be getting out of it and what he’s experiencing just makes the whole thing worse.
I think the issue is that he’s approaching it as work. I started having fun with it when I stopped thinking of it as having limitations.
Electronics, of all things
I’ve been working with IoT and AI for about 15 years, and I’ve never managed to get electronics to work. Bought the starter kits, watched the tutorials, wired things up on breadboards. Never got anywhere. I have a drawer full of components I bought with good intentions, and every one of them ended the same way: I’d hit a wall, not know how to get past it, and move on to something else.
At the beginning of March I pulled all of that out again, because I wanted to monitor my seedlings and I figured Claude could probably help me get past the walls this time. Not for work, just because I wanted to see if it would grow.
The first weekend I was just following instructions. Claude would say “connect pin 27 to the signal wire on the DHT22” and I’d do it, not really understanding what I was doing or why. I broke a PIR sensor because I wasn’t confident enough to catch a mistake before powering it on. The following week, I asked Claude to make me an interactive wiring diagram instead. It generated an HTML playground where I could click on whatever wire I was working on and it would highlight both ends. I took photos of my actual desk and had Claude rearrange the layout so the diagram matched what I was looking at in the real world. After a couple of rounds, it was a perfect match. I just followed the colours.
And every time something worked, I realised I could push further than I thought. The wiring worked, so I pushed to firmware. Claude wrote a dark-themed dashboard on the little screen, a web interface I could check from my phone, over-the-air updates. It worked on the first flash. So naturally, I pushed further. Can we make a proper circuit board? I’ve worked in IoT long enough to know what a PCB is, I just can’t design one.
When I first asked Claude if it could help me design a PCB, it very helpfully told me I’d need to do that myself and explained how. I pushed back, told it to go see what it could figure out, and it came back with tscircuit, a tool that lets you define a circuit board in code and export manufacturing files. Even Claude doesn’t know what Claude can do. TypeScript becomes a physical object. I still find that a little wild.
Three weeks after pulling my old components out of a drawer, I’ve ordered custom PCBs from a factory in China. They cost $2 each. I’m not entirely sure they’ll fit, because the measurement tools I had available were “take a photo and ask Claude to count breadboard rows.” If the measurements are right, I’ll have a proper plant monitor for the garden. If they’re wrong, I’ll have five small green coasters.
It’s not just tech
My wife paints watercolours, and sometimes she’ll be working on something and feel like it’s not quite right without being able to say why. She takes a photo, sends it to Claude, and gets feedback on composition, colour balance, where the eye is being drawn. She’s not using AI to paint. She’s using it the way you’d use a second pair of eyes from someone who knows what to look for.
Neither of us is doing anything we’d call “prompting.” We’re doing the thing people have always done: looking at something, giving or getting feedback, adjusting, trying again. Describing what you want and working with someone to get there. The “someone” just got a lot more capable.
It’s not prompting
If you’ve tried AI and it didn’t click, that’s more common than the hype suggests. A lot of people walk away frustrated, feeling like they’re missing something everyone else seems to get. That frustration often comes from thinking about it as prompting, as if there’s a correct way to ask and you just haven’t found it yet.
But that’s not how any of this worked for me. I’m building a plant monitor because it sounded fun, and then every wall I hit turned out to be shorter than I expected. My wife isn’t optimising her watercolours, she’s getting better at something she already loves. Neither of us sat down and learned to prompt. We just started talking to the tools the same way you’d describe to a contractor what you want your kitchen to look like, or give feedback to a developer on your team. You know the outcome, they know the craft. You describe, they build, you adjust.
Find something you think is fun, and point the tools at that. Next week I’ll find out if I have a plant monitor for the garden or five small green coasters. Either way, after fifteen years of buying starter kits and giving up, I finally built my own electronics.