AI got the app working.
What do you check before real users?

Start above the code.

Can a new user understand the product and complete the main task? Vibe-coded products can accumulate screens and features quickly because adding one more thing feels cheap. That makes scope and coherence especially important.

Then inspect the foundations that AI can make look finished.

• Authentication and permissions
• Where secrets and API keys live
• Database access rules and backups
• Payments and subscription state
• Uploads and user-generated content
• Error handling when an external API fails

If the product handles sensitive data, money, health information, regulated workflows, or meaningful security risk, bring in the appropriate specialist rather than treating a general product review as a formal security audit.

Watch for prompt-built UX debt.

AI can generate plausible screens without understanding the hierarchy of your product. Look for duplicated ways to do the same thing, labels that changed between iterations, inconsistent navigation, dead-end states, and screens that exist because you asked for them once rather than because users need them.

Do not keep prompting blindly.

When the app starts feeling tangled, more prompts can turn uncertainty into more surface area. Stop. Write down the actual problem. Decide whether it is product, UX, implementation, or scope. Then make the smallest change that resolves that specific issue.