I've watched this demo more times than I can count, from half the companies in the category, and it lands in the same place every time. Somebody types a sentence into a box. Code scrolls for thirty seconds. A page appears with a hero image, a booking form and three pricing tiers, and the video cuts to applause.

Freeze it right there. That cut is the moment the whole category decided it had won.

The typing was never the hard part

What got solved is real, and I'm not going to be snide about it. Producing a working full-stack app from one sentence is a genuine result, and anyone who burned a week in 2019 hand-wiring a signup form to a Postgres table knows how much time just evaporated. But think about why that half fell first. Code has more training data behind it than almost anything people have ever written down. It compiles or it doesn't, so a machine can grade its own homework in a loop, over and over, with no customer anywhere near the process. Endless examples, objective scoring, instant feedback. Of course it went first.

The other half has none of those properties. The other half is a person. Call her Dana, hypothetically: she runs a mobile dog grooming van, she fills in your beautiful form at 8:40 on a Tuesday night because her golden retriever rolled in something behind the shed, and then she waits.

So you own a very nice table of strangers

Assume the build went perfectly. Real database, real logins, deployed, custom domain, no console errors, the thing genuinely works. Dana's request lands as a row. Now what?

A human being has to notice it. That's the entire back half of this category, and it has been quietly handed back to the owner, who at that moment is holding a wet spaniel. The row has to turn into three things the demo never shows:

  • Someone who picks up when Dana rings the number on the site instead of filling in the form, which is what most people do anyway.
  • A call back that reaches her before she gets to the next groomer on the list.
  • A Thursday morning slot with her name in it, so she stops being a row at all.

None of that is a code generation problem. It's a staffing problem, and generated apps don't come with staff. A 411 Locals study of 85 businesses found 62 percent of calls to small businesses go unanswered, and the Numa Small Business Phone Report puts the share of callers who reach voicemail and never call back at 85 percent. Put those two numbers side by side and the shape of the thing gets uncomfortable. So far the contribution here has been one more inbox, handed to a business that already can't get through the inboxes it has.

What the builders actually do, plainly

v0 generates React components and pages you take into your own project. With Bolt and Lovable, the app gets assembled in the browser and you leave with a deploy button and a repo. Replit runs the whole thing in a container you can share a link to. Those are the descriptions, no adjectives attached.

Not one of them answers a phone. That was never in their scope, which would be fine if the marketing around all of it hadn't drifted into implying that an app and a business are the same object. Your app records that somebody wanted something at 8:40 on a Tuesday. What happens in the ninety minutes after is a separate question with a separate answer, and nobody generated that part for you.

Quick tangent, then I'll come back. Picture the counter of an old tire shop: a metal spike with paper slips jammed down onto it. Every slip gets called back before the guy at the register locks up for the night. No exceptions. That arrangement would out-convert plenty of CRMs, and the interesting part is the rule about the door, not the spike. Software has spent years rebuilding the spike, prettier every time, and skipping the rule.

We built the second half first, on purpose

Rocketship started at the phone, because that's where the money leaks out. An AI receptionist answers your business number around the clock, from the number your customers already have in their contacts. It knows your hours, your services and your prices. It answers what people actually ask, books into your Google Calendar, takes down the caller's details, and hands off to a human when the call needs one.

Then the workers. Sophie emails from your own Gmail address and places live outbound calls in a voice that doesn't announce itself as software. Claire keeps the calendar straight. Max handles invitations. They run on a schedule, and they check with you before they spend money or contact anybody new, which is a constraint I'd argue for even if nobody had asked for it.

The app builder came last, and it does what the category does: describe your business in plain English, get a full-stack app with a customer database, user logins, Stripe payments through your own Stripe account, an admin dashboard, live on your domain in minutes. The difference is what the form is wired into. Dana submits it, that creates a lead, and the voice worker calls that lead. Same record, same system. No CSV export, no Zapier tax, no second dashboard where leads sit and go stale. There's a free tier and paid plans start at 12 dollars a month, so you can point your own number at it and hear for yourself whether the calls get answered.

Attaching a form to an app is easy now. Everybody can do it. Attaching a phone call to the form is the part that took the work, and it's the part the rest of the category keeps calling out of scope.