No-code earned its reputation. Bubble, Lovable, Replit and the rest let one person prove an idea in a fortnight that used to need a team and a quarter. We migrate stalled no-code builds every month, which means we see the failure mode often and the success stories rarely, so discount our view accordingly. The useful question is not whether no-code works. It is how long it works for, and what the exit costs if you need one.
Three things, and they are not small things.
A week of building beats a month of arguing. If the idea is wrong, you would rather learn that from a no-code prototype than from a finished custom system.
A booking form, an approval queue, a simple client portal. Plenty of these never need to be anything more than what a builder produces.
A founder who can ship changes without waiting on a developer moves faster, and that speed is worth real money in the first year.
The ceiling is rarely about features. It is about these four.
Recursive rules, pricing that depends on three variables, scheduling with capacity constraints. You end up with forty workflow steps standing in for twenty lines of code.
Queries that were instant at a thousand rows are not instant at a hundred thousand, and there is a limit to how far you can tune from inside an editor.
Where the data lives, who can reach it, what the audit trail looks like. Enterprise buyers ask, and the answer has to be about infrastructure you control.
Workload-based pricing means the better the product does, the more the platform takes. Owned infrastructure does not behave that way.
This is the number nobody budgets for, so here it is plainly. Migrating a working no-code app is not a rewrite of the idea, it is a rewrite of the implementation, and the expensive part is almost always the data model rather than the screens. A builder lets you define data loosely, which is a feature while you are exploring and a problem when you need constraints, foreign keys and a migration you can run twice. Expect the move to cost meaningfully more than building the same thing from scratch would have, because somebody has to reverse-engineer the rules out of workflows nobody documented. That is not an argument against starting in no-code. It is an argument for starting there deliberately, with an idea of when you would leave.
Prove the idea in no-code. Keep the data model clean while you do it, even though the platform will not force you to. The moment you have a paying customer whose business depends on the thing working, start planning the move rather than discovering it during an outage. Most of the rescues we get called into are not failures of no-code. They are failures to notice that the thing stopped being a prototype about eight months earlier.
If you are not sure people want it, start in no-code. The cost of proving the idea badly is far lower than the cost of building the wrong thing well. If you already have customers waiting and you know the rules, starting in a real stack skips a migration you would otherwise pay for later.
Two signals. A rule you cannot express without fighting the platform, and a buyer asking questions about your infrastructure that you cannot answer. Either one on its own is a warning. Both together mean the migration is already overdue rather than upcoming.
Almost always, and it is part of the work rather than an extra. We map the platform's data types onto a relational schema with real constraints, run the migration against a copy first, and reconcile the totals with you before anything cuts over.
Sometimes, and we will say so when it is. If the no-code build encodes rules nobody wrote down, reading them out of it is worth doing. If it was abandoned half-finished and the rules live in someone's head anyway, a clean build is often faster and cheaper.
Describe what you’re building, what it has to do and when it needs to be live. You’ll get back the architecture we’d use, the trade-offs, and a realistic timeline, before anything is signed.