AI-generated UI for SaaS admin panels: what works, what breaks
An admin panel is where AI UI tools reveal what they were actually trained on. Consumer-app examples (onboarding flows, feed layouts, marketing pages) don't teach a model to render 14-column tables with conditional-visibility bulk actions. Ask most AI tools for a real admin panel and you get back something visually clean and structurally wrong. This is the specific failure mode teams building SaaS run into most. We've written about the extreme version of the same problem in [why AI design tools fall over on ERP screens](/blog/ai-design-tools-fall-over-erp); admin panels are the milder sibling. Here's what actually works, what breaks, and how to prompt AI tools for admin UI in a way that doesn't waste your afternoon. ## The five things admin panels need that consumer apps don't Every admin panel has some subset of these five requirements. Consumer-app-trained AI tools fold on each one. **Data density.** Real admin screens show 8-20 columns per table, dense enough that a power user can see multiple records without scrolling. Consumer tools default to 4-column layouts because that's what mobile-friendly consumer apps use. **Conditional visibility based on role or state.** "Show this bulk action only if the user has the `pricing_admin` role" or "hide this button if any selected row is archived." Consumer tools treat every button as always-visible; permissions are an afterthought. **Bulk actions.** Select 47 rows, apply the same change to all of them, preview, confirm, undo