I had an idea on a Tuesday. By Wednesday evening, it was a live, installable web app with authentication, a database, geolocation matching, and a public URL anyone could try.
I never opened a code editor.
I just talked to MeDo like I'd talk to a co-founder, and it built the thing.
Try the result here: SkillSwap on MeDo Sandbox (demo login: demo@skillswap.app / demo1234)
What I Built
The app is called SkillSwap. The idea is simple: connect neighbours within a couple of miles for skill exchanges, but with one counter-intuitive rule — you can't actually swap skills until you've met for coffee first.
The skill swap is the excuse. Human connection is the product.
That's the whole thesis. A leaky faucet becomes an introduction. A website redesign becomes a coffee. A Spanish lesson becomes a friendship. The app turns urban loneliness into a feature pipeline.
Functionally, it does what you'd expect:
- Sign up, log in, set up a profile
- Pick the skills you offer and the skills you need
- Share your location to find nearby matches
- Express interest in someone, and if they reciprocate, the coffee meeting unlocks
- Propose a time and place, meet up, both confirm it happened
- Only then does the actual skill swap unlock
There's a one-click "add demo neighbours" button so you can experience the entire flow on your own without needing a friend on the other end.
How MeDo Actually Feels to Use
Here's the part that genuinely surprised me.
I didn't write a spec. I didn't draw wireframes. I opened MeDo and said something close to: "I want to build a neighbourhood skill-exchange app where you have to meet for coffee before you can trade skills."
When I wanted to change something, I just said so. "The hero feels too generic. Make it lean harder into the slowness angle." MeDo rewrote the copy, restructured the layout, and shipped a new build.
The whole experience is conversational. You stay in product-owner mode. MeDo handles the implementation.
What's Genuinely Impressive
Three things stood out to me.
1. It Reasons, It Doesn't Just Generate
When the app started timing out on the live URL but worked perfectly on my local machine, MeDo didn't just throw boilerplate fixes at the wall. It traced the request path through every layer of the platform, formed hypotheses, tested them, and eventually identified a subtle interaction between the platform's request handling and the dev server's proxy.
Then it fixed it in three coordinated layers — a custom proxy middleware, a request-format fallback, and a self-cleaning service worker — and explained each one in language a non-engineer could follow.
I wasn't watching autocomplete. I was watching debugging.
2. It Publishes for You
There is no deploy step. No CI configuration. No environment-variable fiddling. When MeDo finishes a feature, it's already live at a public URL. I sent that URL to friends. They sent screenshots back. The loop from idea to "real people using your thing" is shorter than I have ever seen it.
The Honest Limits
I want to be straightforward here, because the AI builder space is full of breathless takes.
MeDo gives you 300 free credits to start. That's enough to build a respectable prototype. Auth, a few pages, a database, a working flow. For testing an idea or putting together a demo, it's plenty.
For anything beyond a prototype, you'll need more. Real polish — responsive edge cases, accessibility passes, security hardening, performance tuning — burns credits faster than the initial build. That's not a MeDo problem; that's just the nature of software. The first 80% is fast everywhere. The last 20% is where the cost is.
It hits walls. Like every AI builder I've used, MeDo occasionally gets stuck in loops. It will try the same fix twice and not notice. It will misread an error message. It will confidently break something that was working. When that happens, you either rephrase the request, roll back, or — eventually — you hand the codebase to a human engineer who can step outside the AI's frame of reference.
Production-grade security is a different game. MeDo writes sensible, modern code. Auth done with proper hashing, parameterized queries, sane defaults. But "sensible defaults" and "this app holds millions of users' personal data" are different bars. For anything sensitive, you'd want a security review by someone who isn't the one who wrote the code.
None of this is unique to MeDo. It's the honest reality of every AI builder right now. The story is "incredible for prototyping, mid for production, requires human engineers at scale." That's not a flaw, that's the current frontier.
Where MeDo Genuinely Wins
Even with those limits, I came away with a clear conviction: this is the best tool I've used for testing an idea.
If you have a concept and you want to know whether it's worth building, MeDo gets you to "real people, real link, real feedback" in hours instead of weeks. That's transformative. Most ideas die in the gap between "I think this might work" and "I have something to show people." MeDo closes that gap.
For founders, that means cheaper experiments. For designers, that means working prototypes instead of static mockups. For non-technical builders, that means actually shipping the thing in your head.
I'd happily build my next prototype on MeDo. And the one after that. When I find something that actually has traction, I'll bring in a human team to harden it. But for that frenzied, exploratory phase where you're trying to figure out whether the idea even has legs — there's nothing else like it.
Try It Yourself
If you want to see what came out of this experiment, here's the live app:
Demo account:
- Email:
demo@skillswap.app - Password:
demo1234
And if you want to try MeDo for yourself, head to medo.dev. Bring an idea. Don't bring a spec. See what happens.
If you build something with MeDo, drop a link in the comments — I'd love to see what other people are shipping with it.
Top comments (0)