AI-built app cleanup: For apps built with Lovable, Bolt, Cursor or Claude Code
For founders and teams whose app was built fast with AI tools and now mostly works, and feels like several apps stitched together. I take an inventory of what the app actually ships, set one design system your coding agent follows, and fix the flows that lose people. What this page adds to a design system is the triage that comes first: what to keep, what to fix, and what is engineering. What it covers depends on the app.
To start, email andriy.dsgn@gmail.com with a link to the app and a line about what feels off.
What usually goes wrong
An agent is good at one screen. Across a product it repeats whatever it finds, mistakes included, and when nothing answers its question, it invents a value. On a test app an agent built in one evening, the inventory found 21 button variants across 25 screens, three of them defects. Each screen looked fine on its own.
- Several button styles and more than one way to confirm an action.
- Flows nobody designed end to end: the screen exists, the path between screens does not.
- Spacing, colours and type that drift a little on every new page.
- A product that looks like every other app built from the same starter kit.
Developer or designer first
If the app loses data, breaks sign-in or is not safe, that is engineering, and it comes first. Beyond that, product and engineering questions are often settled together: a product decision can remove a part that would otherwise need rebuilding. If the app works and people get lost, that is design. How to tell: Developer or designer for an AI-built app.
How the cleanup works
- Inventory. Every component and value the app ships, from the running product, not from the code alone.
- Triage. What to keep, fix or rebuild, and whether each fix is design or engineering. What it covers, and how it is written up, depends on the app.
- One system. Tokens and components in code, each described for people and for agents, and a rules file your agent reads, tested on real tasks.
- Flows. The flows that lose people, redesigned end to end.
- Handover. One change per task, so your team or your agent can work through them.
What the system needs so agents follow it, measured on 11 builds: Design systems for coding agents.
What it costs
| Step | Price | Time |
|---|---|---|
| Consistency Review | From $700 | 3-5 working days |
| Design system, repaired or built | From $8,000 | Usually about a month |
| A redesigned flow | Quoted per flow | In the flow list |
| Rebuilding the first version | From $8,000 | In the quote |
Every step is priced before it starts. The fee for a Consistency Review or a Product Diagnosis is credited in full toward the work that follows, if it starts within 30 days. That work is quoted separately.
Where it comes from
On Fully AI, an AI agent product, the work was a UX audit, then a design system and a dashboard redesign, in one month. The method above comes from my own experiments with coding agents, written up with the numbers.
Questions founders ask
Can I keep building with Lovable or Cursor afterward?
Yes, and you should. The point of the system is that the next screen your agent builds has the same rules to follow, and a check that shows when it does not.
Will you rewrite my app?
Not by default. The review shows what can stay. Often the components are corrected once and the screens follow.
Do you need access to the code?
For the review, a link to the running app is enough. For fixes in code, access to the repository.
How much does it cost?
The Consistency Review starts at $700. The system after it starts at $8,000, and redesigned flows are quoted one by one after it.
How long does a cleanup take?
The review and triage take 3-5 working days. The work after it is quoted with dates, step by step.
Can you fix it in the code, or only in Figma?
In the code as well. I am a product designer and design engineer, so I can change components and screens in your repository myself.
Which AI builders do you work with?
Lovable, Bolt, Cursor, Claude Code and v0, and apps built by hand. Anything that runs in a browser can be reviewed.
What if the problem is the back end?
Then I say so in the review, and you need a developer first. If you want, I can also rebuild the first version myself.
Other ways I can help
- Design systems, from $700: a Consistency Review, then a system agents follow.
- SaaS UX redesign, quoted per flow: for products that are powerful and confusing.
- Design and build, from $8,000: a first version designed, and built if you want.
Further reading: Developer or designer for an AI-built app. All services, with the situation each one fits: Services.
Built with AI and feels off?
Send a link and a line about what feels off. If it’s engineering, I’ll say so.