Developer or designer for an AI-built app: How to tell which problem you have
If an app built with Lovable, Bolt, Cursor or Claude Code breaks, loses data, runs slowly or leaks, start with a developer. If it works but people get lost in it, every screen looks slightly different, or nobody can say what to build next, start with a designer, ideally one who can read and change the code. Most AI-built apps have some of both. Urgent engineering problems come first; much of the rest is decided together, because a product decision can remove a part that would otherwise be rebuilt.
Symptoms and who fixes them
| What you see | Usually | First step |
|---|---|---|
| Errors, crashes, actions that fail without a message | Engineering | A developer reads the code and the logs |
| Sign-in, payments or user data you do not trust | Engineering and security | A developer, and a security review before real users |
| Pages that slow down once there is real data | Engineering | A developer profiles the queries |
| Every screen looks slightly different: buttons, spacing, colours | Design system | An inventory of what the app ships, then one system |
| People sign up and never reach the point of the product | Product and UX | A designer walks the first-run flow |
| Features keep growing and nobody can say which ones matter | Product | A product audit: what to keep, cut and change |
| It works, but looks like every other AI-built app | Design | A designer, with your brand and a system the agent can follow |
What goes wrong in AI-built interfaces
In a review I ran on an app an agent had built, 25 screens carried 21 button variants: eight were intended and three were defects. The inventory that told them apart: A canvas for reviewing AI-built UI. In another experiment one agent built the same product 11 times, with no system, on shadcn/ui and on a product with its own look: Design systems for coding agents.
The pattern is the same each time. Every screen follows the prompt that made it, so the product drifts unless something outside the prompts holds the rules.
What founders post about it
In public posts about AI-built products that I read in November 2026, most of the trouble was engineering. Of ten posts about apps built with Lovable, Replit, Base44, Claude or Codex, seven were about the back end, finishing the product, scale or hiring a developer, two were about the interface alone, and one was about both. It is a small sample. It matches what founders describe: the first wall is usually the engine, and the second, once the app works, is that nobody designed it as one product.
If you need both
- Fix what is urgent first: lost data, broken sign-in, anything that touches money or is not safe.
- Settle the product question alongside the remaining engineering: what the app is for and what to cut. The answer can remove parts that would otherwise be rebuilt.
- Then the interface, with one design system your agent can follow, and a check that shows when new screens drift.
- Keep prompting the AI tool for small changes once the system exists. It is good at that.
Where I fit
I am a product designer and design engineer, so I work on the design side and can change the code myself. For an AI-built app that means a Consistency Review, from $700; a Product Diagnosis, from $900, when the question is what to build; a redesign, flow by flow; or design and build when the first version has to be rebuilt. When the problem is engineering, I say so.
Questions founders ask
Should I hire a developer or a designer for my Lovable app?
A developer if it breaks, loses data, is slow or is not safe. A designer if it works but people get lost, the screens do not match, or nobody knows what to build next. If both, fix what is urgent first and decide the rest together.
Can I fix the design by prompting Lovable or Cursor?
Small, local changes, yes. A product with no shared rules, no: each new screen follows its own prompt. A design system written for agents gives every new screen the same rules to follow, and a check of what ships shows when one does not.
Do I need a design system for an AI-built app?
Some shared rules, yes, once more than one person or agent adds screens. That can be a small set of tokens and components rather than a large project; the size follows how far the product has drifted.
How much does it cost to fix an AI-built app’s UI?
With me, a Consistency Review starts at $700, and work on the system after it at $8,000. Redesign is priced per flow, after the review.
Will a designer rewrite my app?
Not by default. The review shows what can stay. Often the components are corrected once and the screens follow.
Not sure which problem your app has?
Send a link and a line about what feels wrong. If it’s engineering, I’ll say so.