September 28, 2026
Stuck at 80%: What to Do When Your AI-Built App Stops Working
Before you pay anyone to fix it, find out whether anyone will pay for it.
The short answer
Don’t fix it yet. Sell it first. If your Lovable, Bolt, Replit or Bubble app is stuck at 80% and you have zero paying customers, the next dollar you spend should go on finding out whether a business will pay for the outcome, not on code. Offer a small paid pilot you can deliver partly by hand. If buyers say yes, fix or rebuild only what that pilot needs. If they say no, you’ve just saved the cost of the fix.
Why does my AI-built app keep breaking at 80%?
It breaks because the last 20% is the part AI tools are worst at and non-technical founders can least see: logins, payments, permissions, data and the way all the pieces talk to each other. The first 80% (screens, buttons, a demo that looks real) is fast. The last 20% is where one fix quietly breaks two other things.
Kaia Colban described exactly this on Indie Hackers after Cursor added a second login system that broke the first:
“I couldn’t read the code well enough to see that and properly instruct Cursor…”Kaia Colban, Indie Hackers, 6 September 2025
That’s the trap. You can’t read the code, so you can’t tell the AI what’s actually wrong, so you describe the symptom again, and it tries another guess. A few rounds later you’re on your third rebuild.
You are not alone in this. Lovable’s own 2026 Build Economy survey of more than 14,300 users found:
These figures are self-reported by Lovable’s users and drawn from its own platform. The Next Web noted they have not been independently audited.
Read that last number again. Most people building these apps plan to make money from them and haven’t yet. The code is only half the problem. The other half is a distribution problem: nobody has been asked to pay.
Should I fix it, rebuild it or sell it first?
Sell it first unless you already have paying customers or signed buyers waiting. Use this table to find your path honestly:
| Path | Choose it when | Warning signs you’re on the wrong path | Your first move |
|---|---|---|---|
| Sell it first | Zero paying customers. No buyer has said “yes, here’s the money” for this exact outcome. You’re about to spend on a fix. | You keep saying “once it works, I’ll start selling”. You’ve shown friends, not buyers. | Take the Manual MVP Check, then offer a paid pilot you can deliver partly by hand. |
| Fix it | Buyers have paid or signed a pilot, and the problems are specific (one feature, a login bug, a slow page). The core of the app works. | Every fix breaks something new. You can’t list what’s broken in one page. | Pay for a short code audit and get a written list of fixes before agreeing to any bigger quote. |
| Rebuild it | Paying customers exist, but the foundations are wrong: data mixed between customers, no way to secure it, or it can’t handle what buyers already pay for. | You want to rebuild because it “feels messy”, not because a paying customer needs it. | Write down what the paying pilot actually needs, then get two quotes scoped to only that. |
Notice that fixing and rebuilding both start with paying customers. Without them, you’re paying to perfect a guess.
What does it cost to fix an AI-built app?
Fixing a stuck AI-built app typically costs anywhere from your own time to tens of thousands of dollars, depending on who does it. One published guide from BigIdeasDB puts taking a broken vibe-coded project to launch at US$2,500 to US$9,000, starting with a paid audit. Here are the realistic options side by side.
| Option | Typical cost | Good for | The catch |
|---|---|---|---|
| DIY with the AI tool | Your subscription or credits, plus your time | Small, clearly described bugs | If you can’t read the code, you’re guessing. This is how third rebuilds happen. |
| Freelancer | US$50 to US$150 an hour; roughly 10 to 20 hours for simple cleanup, 40 to 80 hours for a full refactor (BigIdeasDB) | A specific list of fixes after an audit | Quality varies a lot. You can’t judge their work, so scope creep is easy. |
| Rescue service | Full audit and report around US$1,500; full-stack rescue US$3,000 to US$8,000 (vibecoding.app) | Security holes, broken logins, getting to “launchable” | They fix code. They don’t tell you whether anyone will buy. |
| Australian agency rebuild | Simple MVPs from about A$40,000; mid-sized agency rates A$150 to A$330 an hour (Conduct) | A proper rebuild once buyers are paying | Expensive for something unproven, and quotes for new products tend to grow. |
| Fractional CTO | A$5,000 to A$15,000+ a month on retainer (AI Talent on Demand) | Ongoing technical leadership once revenue justifies it | A big monthly commitment with zero paying customers. |
Prices are published ranges from the sources listed below, not quotes. US$ figures come from US-based sources; A$ figures from Australian ones. Your real cost depends on what’s broken.
If both quotes feel wrong, that’s usually because you’re being asked to price a fix without knowing what the fix is for. A quote tied to “make it work” has no edge. A quote tied to “make this one pilot deliverable for this paying customer” does.
Should I keep using Lovable or Bolt, or hire a developer?
Keep using the AI tool for as long as it’s helping you sell, and hire a developer only when a paying customer needs something the tool can’t deliver safely. Lovable, Bolt and Replit are excellent for demos, clickable prototypes and internal tools. They are much weaker at the invisible parts B2B buyers care about: security, data separation between customers and reliability.
So the question isn’t “tool or developer?” It’s “what does my first paying customer need, and can I deliver it with what I have, plus some manual work behind the scenes?” Often the answer is yes.
How do I check if anyone will pay before I fix it?
Ask for money for the outcome, not the app, and deliver the first few by hand. This is the idea behind a manual MVP: you, a one-page offer, a payment link and AI tools doing the heavy lifting behind the scenes. Your 80% prototype can still play a part as a demo. It just doesn’t have to carry a live customer yet.
- Write the outcome in one sentence. What result does a business get, by when? Not “an AI dashboard”. Something like “a weekly compliance report, done for you”.
- List 20 real businesses you can reach this month. Use your 10+ years in the industry. People you’ve worked with, or people like them.
- Show the demo, then ask for a paid pilot. A small, time-boxed engagement with a clear price. A “maybe later” is a no for now.
- Deliver it partly by hand. Use the working parts of your app, plus spreadsheets, email and AI for the rest.
- Only then decide: fix, rebuild or walk away. Now you know exactly what a paying customer needs, which is the brief any developer should be quoting against.
Not every idea can be delivered by hand. The free Manual MVP Check takes two minutes, gives you a green, amber or red score, and shows what works by hand in your sector.
Is my AI-built app safe to put in front of customers?
Assume it isn’t until someone has checked. AI tools can produce apps with logins that can be bypassed, customer data that other customers can see, or secret keys left in the code. For a B2B buyer, that’s a deal-breaker, and it can create legal exposure for you.
If you plan to let a paying customer use the app itself during your pilot, check it first. The guide Is your AI-built app safe to sell? walks through what to look for, and the Is My AI MVP Safe to Sell? workshop (US$67) takes you through it live.
Disclaimer: this is general information, not legal or security advice. Always check the local laws and regulations on privacy, data and consumer protection in your applicable market(s) before you sell or deliver anything.
Before you pay anyone to fix it, check who actually holds the keys: the code, the accounts and the data. It’s free, and it takes no coding.
Free founder audit
Who really owns your AI-built app?
A 20-point audit for founders with a Lovable, Bolt, Replit or Cursor prototype, or a freelancer using AI. No coding needed: you're checking who holds the keys, not reading code.
Who is this approach not for?
If you just want someone to fix your code, that isn’t what The Lean CPTO does, and that’s completely fine. A good freelancer or rescue service from the table above is the right call, especially if you already have paying customers. Ask for a short audit first, and keep ownership of your code, data and accounts.
This approach is for you if you’re a non-technical B2B founder with deep industry experience, a prototype that’s “nearly there” and zero paying customers, and you’d rather get your first paid pilot than your next rebuild.
Will anyone pay for it?
Find out before you pay to fix it. Take the free Manual MVP Check: 5 questions, 2 minutes. You’ll score green, amber or red, then see what works by hand in your sector. Or work through fix, rebuild or pause on your own app, live, in the Stuck at 80% workshop.
Get the free Manual MVP Check Join the Stuck at 80% workshop (US$67)Frequently asked questions
Why does my Lovable app keep breaking?
Most breakages happen in the last 20%: logins, payments, permissions and data. AI tools fix one symptom at a time, and if you can’t read the code, you can’t tell them what’s really wrong, so each fix can break something else.
Should I fix or rebuild my AI-built app?
Neither, until someone has paid. If you have zero paying customers, sell a small paid pilot first. Fix the app if paying customers need specific repairs. Rebuild only if paying customers need something the current foundations can’t support.
How much does it cost to fix a vibe-coded app?
Published ranges vary widely. One guide puts a broken vibe-coded project at US$2,500 to US$9,000 to fix and launch. Freelancers often charge US$50 to US$150 an hour, and rescue services around US$3,000 to US$8,000. Start with a paid audit so you know what you’re fixing.
Is it better to keep using Bolt or hire a developer?
Keep using the AI tool while it helps you sell and demo. Hire a developer when a paying customer needs something the tool can’t deliver safely, such as proper security or data separation between customers.
Can I sell an app that’s only 80% done?
You can sell the outcome. Offer a paid pilot, use the parts of the app that work, and deliver the rest by hand with spreadsheets, email and AI tools. Check that anything customers touch directly is safe first.
Related guides
- The Manual MVP: get paid before you write code
- Is your AI-built app safe to sell?
- What an MVP should really cost in 2026
Sources: Lovable, The Build Economy data study (2026) (non-technical, solo and monetisation figures; survey of 14,300+ users, May 2026), The Next Web (self-reported, not independently audited), Indie Hackers, Kaia Colban (quote), BigIdeasDB (US$2,500 to US$9,000 fix and launch; freelancer rates and hours), vibecoding.app (audit and rescue pricing), Conduct (Australian agency costs and rates), AI Talent on Demand (Australian fractional CTO retainers).
About the author
Matt Ainsworth is the founder of The Lean CPTO. He has spent 20+ years across product, technology, sales, marketing and communications in Australia and Japan, including nearly a decade in Tokyo, and more than 10 years mentoring founders. He has worked on both sides of the developer relationship: selling development services and building products with developers.