Tools such as Lovable and Replit have made coding vastly accessible to non-technical builders, but the result is a sea of cookie-cutter apps that are hard to distinguish. Paul Bakaus, CEO of AI design startup Impeccable, noted in a June 23 podcast interview with Andreessen Horowitz that these AI-designed apps share distinct visual flaws.
The first telltale sign is what Bakaus calls "regression to the mean" — a painfully mid design that is boring and cookie-cutter. These apps converge into a beige, sans-serif haze, looking similar and lacking originality. While they may function adequately at a small scale, the devil is in the details.
A second red flag is dysfunction masked by polish. AI-generated apps often feature pretty layouts that break under real-world use, leading to usability problems that become glaring as the app attracts more users. The third sign is a lack of cohesive user experience — design choices that work in isolation but fail to form a seamless whole.
These issues become critical when apps go commercial. An app that looks AI-coded may struggle to build trust with users or investors, and scaling can expose underlying design flaws that lead to poor retention. Non-technical founders relying on vibe coding risk building products that are hard to iterate on without deep design expertise.
Bakaus did not offer a remedy in the interview excerpt, but customization and human-led design refinement are widely seen as necessary steps to stand out. The broader implication: while AI democratizes creation, it may not yet replace the nuance of professional design.