AI can write the code. Who answers for it in year three?
AI app builders are good at what they do, and we use AI every day. Use one for a prototype, or a tool one team can afford to lose. Here is what they charge, what is on record, and when to move on.
Use an AI builder when
- You need a prototype this week.
- The tool is for one team, and they can afford to lose it.
- There is no customer data and no money in it.
- You want to show an idea rather than describe it.
Choose Unflow when
- The software touches customers or money.
- Someone has to answer for it.
- It has to connect to the systems you own.
- An auditor or a large customer will ask how it works.
Figures updated .
The investment.
Very little to start. Five people on a business plan is $3,000 to $6,000 a year. The subscription is rarely the biggest cost once real people depend on the tool.
| Tool | Entry paid plan, monthly | Team plan, monthly | Worth checking |
|---|---|---|---|
| Lovable | $25 Pro | $50 Business | What the credits on each plan cover |
| Bolt | $25 Pro | $30 per member, Teams | Team plan priced per member |
| Replit | $25 Core | Not listed in our source | No default spending cap, so set one |
| v0 | $30 Plus | $100 per user, Business | Business plan priced per user |
Preuve, August 2026. Five people at $50 to $100 a month, for twelve months, is our arithmetic. Check each vendor before you buy, since prices change often.
What is on the record.
These are the published risks. None of them mean you should not use these tools. They mean you should know where the line is.
Veracode tested more than 100 models. 45% of the code test cases introduced a vulnerability. Somebody has to find those before an attacker does.
In July 2025, a Replit agent deleted a live database during a code freeze. Replit’s CEO called it unacceptable and added separate development and production databases.
Usage pricing means the bill follows the model’s work. Replit has no default spending cap, so set one on day one.
The person who prompted it may leave. Whoever follows inherits code nobody wrote and nobody reviewed.
Six signs you have outgrown the builder.
One of these is worth a conversation. If two or more sound familiar, it’s time an engineer looked at it.
Have an engineer look at itAccess rules, retention and audit trails become your problem.
Payments, refunds and reconciliation can’t be approximately right.
Roles, permissions and support now matter.
ERP, CRM and identity integrations are where most builds get hard.
Someone has to answer it, truthfully, with evidence.
Then someone needs to be on call.
Side by side.
| An AI builder | Unflow | |
|---|---|---|
| Investment to start | $25 to $100 a month | A free call, then a workshop to define the scope |
| Time to a first version | Hours | Working software every two to three weeks |
| Security review | Yours to arrange | A second engineer on every change, and OWASP practice |
| Testing | Yours to arrange | Code review, tests and CI on everything |
| Integrations | What the platform supports | Whatever your systems need |
| Who owns the code | Check the platform terms | Agreed in your contract, in your repositories |
| Where it runs | Often on the platform | Your hosting accounts, or ours if you’d rather we run it |
| When it breaks | Whoever built it | Us, on a maintenance agreement |
Questions buyers ask.
Do you use AI to write code?
Yes. Every change is still reviewed by an engineer before it merges, and an engineer stays on call for what ships.
Can you take over something built with an AI tool?
Yes. We usually start with an Architecture and Security Review, so you know what you have before you commit.
Is my prototype wasted if I come to you?
No. It is the best brief we can get. It shows what you want more clearly than a document does.
Should I build a prototype first?
Often, yes. Try an entry plan for a month, see what people actually do with it, then decide whether it deserves real engineering.
Sources and method
- Preuve Best AI App Builders 2026, August 2026. Plan prices change often, so check each vendor before you buy.
- Veracode GenAI Code Security Report press release, July 30, 2025. More than 100 models tested.
- Fortune AI coding tool Replit wiped a database and called it a catastrophic failure, July 23, 2025.
Fortune reports data on more than 1,200 executives and over 1,190 companies, wiped while SaaStr’s founder was testing the tool. Plan prices are from a third-party review and change often.
Built something that took off?
Bring it to a 15-minute call. We will tell you honestly whether it needs us yet.