On software for a business, built and run by agents: what works, what breaks, and the numbers behind it.
Most AI app builders are judged by the first screen they draw. A business runs on what happens after: the data, the clients who sign in, the work at nine every morning, and who fixes it when it breaks.
Seven options compared on the bill, the hosting, the jobs and the data, with our own count of how many Replit apps were still up a year later.
Airtable has a new owner. Eight alternatives on price, limits and client access, and every way to build an Airtable client portal, with what it costs.
One rule — a cell divides once it has eaten enough — grows a fractal. Point the same rule at a function and you get a famous optimisation algorithm, which we benchmarked until it broke. Both simulations run live on the page.
The preview is a false green. Data, secrets, auth and integrations all pass in the sandbox and fail on deploy. The gap, charted — and what it takes to actually ship.
We modeled how token, credit, and effort pricing scale with codebase size across Bolt, Lovable, v0 and Replit. The math points to one wall — and how to design around it.
One generates gorgeous React components you wire up and ship on Vercel. The other hands you a deployed app. That gap is the whole decision.
One drops you into a browser IDE with a token meter running. The other hands you a deployed app. That gap is the whole decision.
Both turn a prompt into a real app. The difference is what happens after “build” — who owns the code, who deploys it, and what the credits actually cost.
Chess looks like a beginner project until you try to write the rules. I gave an AI builder two words — and here's where “easy” quietly ends.
AI app builders solved how to ship in ten minutes. They did not solve how to be found. There's a fix, and it takes about an hour.