In the last year, solo, I've shipped:
- An MCP security gateway with policy enforcement and audit logs
- An npm package of ~20 MCP tools doing ~1,600 weekly downloads
- A live market-stats widget for real estate agents
- A listing binder generator for agents (in beta)
- A 27,000-word technical course across seven modules
- Plus a few things I killed on the way
Ten years ago, that's a fifteen-person company with a seed round. Today, it's one person with a laptop, a Claude subscription, and unreasonable working hours.
That's the story everyone's telling about AI right now. It democratizes building. Anyone can ship. You've read the take. I'm not going to just repeat it, because the honest version is more interesting than the hype version.
The tension nobody wants to sit in
Most "AI gives power back to the people" pieces skip the other half of the story: AI also concentrates power in a way we've never seen. Three or four labs control the frontier models. Compute is a moat. Training data is a moat. Distribution on the consumer side is becoming a moat.
Zoom out and this looks like the least democratic technology shift in a generation.
And yet — the tools those labs build have, as a side effect, put more leverage in the hands of individual builders than anything since the browser. Both things are true. Anyone telling you only half is selling something.😉
The useful framing: the platforms are centralized. The leverage is distributed. What you do with that leverage is up to you.
Building wasn't the gate. Shipping was.
Here's what I think people get wrong when they say "AI lets anyone build software."
Anyone could learn to build software before. The tutorials were free. The languages were free. Stack Overflow was free. If you wanted to write a web app in 2015, nothing was stopping you except time.
What was actually expensive was the distance between "idea" and "live product someone uses." That distance included: learning three frameworks you don't care about, configuring infrastructure, debugging auth flows at midnight, writing the eighty boring lines between the two hundred interesting ones, designing a landing page, setting up billing.
Every one of those friction points was a tax on trying. And because trying was expensive, most people didn't. They had the idea, weighed the six weekends it would cost, and shelved it.
What AI actually changed isn't capability. It's the marginal cost of an attempt.
I had an idea for a hosted stock analysis tool. I built it, shipped it, realized the liability exposure was too ugly, killed it. Total cost: a weekend. I had an idea for a real estate CMA product. I spent a few days on it, figured out Texas is a non-disclosure state and the market was already crowded, killed it. Total cost: a few days.
Ten years ago, each of those dead ends costs me a month. I'd have killed maybe one of them. The other I'd have stubbornly shipped and wasted six more months on.
Cheap attempts change what's rational to try.
What's actually scarce now
If building is cheap and shipping is cheap, the honest question is: what's the new bottleneck?
Three things, in my experience:
Taste. When you can build anything in a weekend, choosing what to build matters way more than being able to build it. Most AI-generated products I see lose not because the code is bad but because nobody asked whether the thing should exist. Taste used to be a nice-to-have on top of hard-won engineering chops. Now it's the whole game.
Distribution. You can generate a product in an afternoon. You cannot generate an audience in an afternoon. The builders winning right now are the ones who were quietly building an audience for years before the tools caught up — or who accept that distribution is now the hardest part of the job and treat it that way.
Persistence. Cheap attempts mean more dead ends, not fewer. The people who win are the ones who can ship something, watch it flop, ship something else the next week, and not spiral. That's a temperament, not a skill.
The quiet implication
"AI gives power back to the people" is the feel-good version of this story. The more accurate version: AI offers power to anyone willing to pick it up.
It's still work. It's still taste. It's still showing up on the eighteenth weekend when the first seventeen didn't pan out. The ceiling has been raised, but the floor is still the floor — you have to stand on it.
The builders who'll matter in this era aren't the ones who can use the new tools. Soon, that's everyone. They're the ones who decide to.
If you've been sitting on an idea, your excuses just got up and walked out.
![Defluffer - reduce token usage 📉 by 45% using this one simple trick! [Earthday challenge]](https://media2.dev.to/dynamic/image/width=1000,height=420,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fiekbgepcutl4jse0sfs0.png)
