AI will happily design the wrong thing for you (www.antonsten.com)

Product designer and author of Products People Actually Want clarifies she isn’t anti-AI — she uses tools like ChatGPT, custom copywriter agents trained on brand voice, Visual Electric, Granola and Cursor to speed work — but warns AI will happily design the wrong thing if human judgment is missing. Lowered barriers make it easy to build things that technically “work” but lack product-market fit or polish; a recent example is a Shopify template that reproduced a food influencer’s cookbook cover, underscoring risks around lazy reuse, IP, and model output that confidently fills gaps with plausible but incorrect details (hallucinations). For the AI/ML community this is a practical reminder: models are leverage, not substitutes for user research, taste, or domain expertise. Key implications include the value of fine-tuning/custom agents to encode brand voice, the rising importance of prompt engineering and human-in-the-loop validation, and the need to detect and prevent inadvertent replication or biased outputs. Designers and engineers should master curation, interpretation of model outputs, and iterative refinement; otherwise AI simply amplifies poor ideas faster. The future favors practitioners who combine human insight with model capabilities to produce intentional, validated products rather than polished—but wrong—automations.