A little over a month ago I wrote about how the design team at Legl had started experimenting with Figma Make for rapid prototyping. It was fast becoming our go-to tool. I even declared: “The AI prototyping race is Figma’s race to lose.”
Well. Here I am, barely a handful of weeks later, eating my own words, as I think the race might already be lost.
From excitement to exasperation
At first, it was thrilling. We wanted to speed things up, so we pulled our design system into Make.
And for a while, it worked. Prompt by prompt, Make spat out React components from our Figma library. But the honeymoon ended quickly. As soon as we tried stitching components into actual screens, it all fell apart. Make defaulted to inventing new components instead of reusing ours. And despite showing the code in an editor window, any manual overwrites were wiped the minute the AI touched that file again. It was incredibly frustrating.
Bolt of lightning ⚡
So we looked elsewhere. Enter Bolt.
To our surprise, it didn’t take much to export the components we’d sweated over in Figma Make and spin them up in a Bolt project. Suddenly, we were flying. Screenshots from our platform became working flows, stitched together with our real components. Entire user journeys built in hours, not days. And the integrations! Supabase for seamless database hookups. GitHub for branching. Netlify for deploy previews. For a moment it felt like we’d finally found the promised land.
But then GitHub integration bit us. Hard.
The GitHub mirage
Bolt dangled the dream of seamless GitHub collaboration, but reality was cruel. There was simply no way to have a private GitHub repo shared between multiple Bolt team members. Sure, we could make the repo public, but no one was comfortable with that. We’d come too far to settle for a half-baked setup.
So, reluctantly, we packed our bags again. And stumbled into a place no one expected.
Cursor: the unexpected destination
Cursor was a surprise because, let’s be honest, it’s built for developers. For designers, it’s terminal-land — a scary place filled with cryptic commands and black screens.
But there was also a guilty thrill. Typing ‘npm install’ and watching the console explode into life made me feel like I’d hacked the Pentagon. The reality was messier – endless failed commands, obscure errors, and painful repetition with each teammate that we wanted to get setup. The same commands that worked for me mysteriously broke for others. It was… a slog.
Yet somehow, we made it.
Running at last
Now we’re setup our terminal commands are reduced to just ‘npm run dev’ and ‘git pull’. I’d prefer to just press a button to do these actions, but we can live with it because now, we are actually running. Cursor, with its team plan and beefy credit allowance, interprets screenshots just as well as Bolt. We’ve got branches, we’ve got deploys, we can merge, we can update. We’re finally in motion.
The best part: nothing’s permanent. If Bolt sorts out GitHub, we can switch back instantly. Or we could split: some designers on Bolt, others on Cursor. Heck, if Figma Make wakes up and catches up, we could even swing back there. The only non-negotiable: we need to access a private GitHub repo where the team can push and pull freely.
How real is too real?
So now the bigger question now is: how far do we take this?
For example, when I wanted to mimic a company search, instead of faking a list of businesses I hooked straight into the Companies House API. It worked beautifully — real data in a prototype. But does it make sense? Engineers will rebuild that feature properly anyway. Yes, it might surface edge cases earlier in the process, but it won’t really save them time.
That’s the balance we’re wrestling with. Prototyping with AI should make us faster, help us test ideas, and validate solutions quicker. The danger is going so deep into “realism” that we’re effectively pre-building production features in a playground.
So every time we’re tempted to wire up another database or API, we stop and ask: is this going to save us time overall? Will it help us design better solutions?
Still muddling through
And that’s where we are: somewhere between exhilaration and exhaustion. We’ve left Figma Make behind, flirted with Bolt, and found ourselves knee-deep in Cursor.
It’s been messy, frustrating, occasionally electrifying — but progress never looks neat from the inside.
The exodus is real. The destination? Still very much TBD.