Practical AI Is Boring. I Think That’s the Point.

I used Figma’s new MCP Server to generate some UI code. It worked. Then it didn’t. AI isn’t magic. It’s a dishwasher. It’s not here to change your life. It’s here to rinse off the boring stuff.
Practical AI Is Boring. I Think That’s the Point.

Today I tested Figma’s freshly-announced MCP Server Beta to help rebuild a button in our design system. The code it gave me was surprisingly decent — clean, close to what I wanted, and almost fully functional. I was impressed enough to pull my coworkers into a meeting to show it off.

But then I tried the same prompt again. It churned for a bit, and… spat out something completely different. This time, the code wasn’t great. Some things I liked were gone. Other things I definitely didn’t ask for were suddenly there. The illusion faded fast, and I was jolted back to my usual AI mindset: this isn't a miracle.

I'm not anit-AI. Not anymore, at least. I was at first and there are still parts I don't like. But like it or not I have to start using it at work. Quite literally, I was told I have to. So, I have. Not with open arms but with an air of healthy skepticism.

That Figma MCP scenario helped re-ground me, and reaffirm a new mental model for AI: it's an appliance.

Not in the flashy, smart-fridge-with-a-touchscreen kind of way. More like… a dishwasher. It does one job. Sometimes it does it well. Sometimes it misses a spot and you have to rewash the pan by hand. But sometimes it saves you time, and that’s the point.

Using AI to help write code, restructure a layout, or draft documentation doesn’t feel like some sci-fi future — it feels like outsourcing a chore I can do, but don’t always want to. I still need to load it properly (give it a prompt), pick the right setting (adjust the tone or code language), and unload it when it’s done (check for wonky edge cases). It’s not magic. It’s mechanical.

And like an appliance, I don’t really care how it works under the hood. I don't think about it while it's running. I care about the results, but only when they're wrong. I’m not putting mission-critical work in its prompt box, but I’ll happily let it scrub away some of the repetitive grime so I can focus on something more interesting.

There’s also a trust line I won't cross. I wouldn’t let my microwave make the Thanksgiving turkey. I wouldn’t ask my dishwasher to dry clean my clothes. I’m not expecting an AI assistant to do strategy, design judgment, or final QA. But I am happy for it to knock out the grunt work that slows me down.

It’s fine. It’s useful. It’s kind of boring. But it’s nice when it makes my day a little easier — and doesn’t leave me scraping off the mess it missed the first time.


P.S. Thinking about doing a follow-up on how I actually use AI day-to-day as a design system product owner — the boring, useful stuff. Let me know if that sounds helpful in the comments or by email.

Or just email me to tell me how dumb I'm being about AI. That sounds fun, too.

Let's talk! Email me — or leave a comment below.

Member discussion