My AI MVP is stuck at 80%: fix it, rebuild it or sell it by hand?
Short answer: you asked it to fix the login. It broke the dashboard. Then it fixed the dashboard and broke payments. You’re not bad at prompting. You’ve hit the wall almost every AI-built MVP hits. There are three ways out: fix it, rebuild it, or pause the build and sell the result by hand. Answer five questions in order and stop at the first match. The first one decides most cases: has anyone paid? If not, pause the build and sell it first.
Why AI-built apps stall at 80%
The first 80% feels fast. Screens appear, buttons work, the demo looks real. The last 20% is where logins, payments, real data and real users live. It’s most of the work, not the finishing touch.
It’s not just you. In the Stack Overflow Developer Survey 2025, 66% of developers who answered said their biggest frustration with AI tools is answers that are “almost right, but not quite”. 45% said debugging AI-generated code is more time-consuming. If professional developers struggle with “almost right”, a non-technical founder or business owner isn’t going to prompt their way out of it.
I spent two years selling offshore development to businesses in Australia and Japan. The projects that got stuck weren’t usually stuck on code. They were stuck on a decision nobody had made.
There are three ways it goes wrong.
Problem 1: you’re finishing an app nobody has paid for
When it’s 80% done, finishing it feels like the obvious next step. But if nobody has paid, or clearly said yes to paying, you don’t know which 20% matters. You might be fixing features nobody will use.
The fix: sell it first. Your customer doesn’t need 100% of your app. They need the part that solves their problem. Everything else can wait, or be done by hand.
Problem 2: you can’t tell a short list from a loop
Some stuck apps need three named fixes. Others are in a loop where every fix breaks something that used to work. From the inside, they feel the same. So people keep prompting, and keep paying.
The AI charges you to create the bug, charges you to fix it, then puts it back.
The fix: prove it. Write down everything that’s broken, and mark each one: blocks a sale, annoys a user, or nice to have. A short, named list can be fixed. A loop can’t be prompted away.
Problem 3: you hand the mess to a developer with no brief
Stuck founders often send a developer a login and “can you fix it?”. Developers quote what you describe. A vague brief gets a vague, expensive quote, and a fix you can’t check.
The fix: build only what’s been paid for. Brief the developer on the few problems that block a sale, with what done looks like for each, and nothing else.
The six ways AI MVPs stall
Tick the ones you recognise.
- The regression loop. Every fix breaks something that used to work.
- The credit spiral. You pay to create the bug, then pay again to fix it.
- Works in preview, not live. Fine in the builder, broken on the real website.
- Half-working integrations. Payments, emails and logins that work some of the time.
- Hidden security gaps. Customer data that other people can see or change.
- Tangled foundations. Code so tangled even the AI loses track of it.
One or two of these, with a short list behind them, is often fixable. Numbers 1 and 6 together usually point to a rebuild. Number 5 always goes first.
Fix, rebuild or pause: five questions
Answer in order. Stop at the first match.
- Has anyone paid, or clearly said yes to paying? No: pause, and sell it by hand.
- Could customer data, logins or payments be exposed? Yes: fix that first, before anything else.
- Does each fix break something else? Yes: rebuild.
- Could a developer understand the code in an hour? No: rebuild.
- Is what’s left a short list of named problems? Yes: fix.
Still unsure? Pause, sell by hand, and get a second opinion before you spend.
If you’re a business owner with a stalled internal tool, swap question 1 for this: is the team using it for real work today, and would they notice if it stopped? If not, pause it and run the job by hand or on tools you already pay for.
What each path means
- Fix it. The core works and the problems are a short, named list. Pay for fixed-price fixes, one at a time. Start small: one paid fix first, before the big job.
- Rebuild it. Every fix breaks something else. Keep the design, the copy and what you learned. Rebuild the core properly. This is the biggest spend, so get two quotes and compare scopes, not just prices.
- Pause and sell it by hand. You don’t know yet if anyone will pay. Stop building, deliver the result manually, and sell first. It costs close to nothing in code: your time, a spreadsheet, email, a form and a payment link.
Whichever you pick, write down one reason why, a budget ceiling and a date to review the call.
If it’s fix or rebuild: a one-page brief
Four things, on one page.
- What it does, and who for. Two sentences a stranger would understand.
- What’s broken. Each problem, with the steps to see it happen.
- What done looks like. How you’ll check each fix yourself.
- Where it lives. The tools and accounts, and who owns them.
Then check the quote before you sign. Is it a fixed price? Is each problem named? Are the code, accounts, logins and data all in your name? Will they leave written notes so the next person can pick it up? A good developer never minds you checking.
If it’s pause: sell it by hand
Customers buy the outcome, not the app. You do the work behind the scenes with email, spreadsheets, forms and AI chat. They pay for the result.
Write down who it’s for, the result they get, how you’ll deliver it by hand this month, your price and how many places. Then offer it to three businesses this week:
“I’m taking on [number] businesses this month to [result]. I’ll do it for you by hand while the software is finished. It’s [price]. Would you like one of the places?”
Every manual step you repeat is a feature someone has already paid for. When paying customers need the same step every week, build that step, nothing else.
Your next seven days
- Day 1: finish your triage and make your call.
- Day 2: send your brief to two developers, or pause the build.
- Day 3: send your hand-delivered offer to the first business.
- Days 4 and 5: follow up and have the conversations.
- Day 6: make at least one clear payment ask.
- Day 7: check every quote before you sign.
FAQ
Can’t I just keep prompting until it works? Sometimes, if the list is short and nothing breaks when you fix it. If every fix breaks something else, more prompts usually means more cost, not more progress.
Is a rebuild throwing away everything I’ve done? No. Keep the design, the copy and everything you learned about your buyers. You’re rebuilding the core, not starting the idea again.
What if I’ve already spent a lot on it? That money is gone either way. The question is what the next dollar buys. If nobody has paid yet, the cheapest next step is a buyer, not a fix.
Should I fix security problems even if nobody has paid? If real customer data, logins or payments could be exposed, yes. Fix that first, or take the app offline until you do.
I’m a business owner and this is an internal tool. Does this apply? Yes. Swap “paid” for “used by the team for real work”. If nobody relies on it yet, pause it and run the job by hand or on tools you already pay for until the same step repeats every week.
The stats in this guide are from the Stack Overflow Developer Survey 2025, linked above.