Two pieces this week are in direct conflict, and nobody has introduced them. Fast Company's piece on giving AI 'good enough' taste describes a world where the floor for AI-generated web design is rising so fast that human designers are being priced out of the low end. The Atlantic's Jenny Odell piece on reclaiming attention describes a human population whose capacity to evaluate anything, AI-generated or otherwise, is degrading at roughly the same pace. These are not separate problems. They are one problem with two symptoms.
The Floor Rises, the Ceiling Disappears
Ben Blumenrose's observation in the Fast Company piece, that AI-generated design's floor is rising fast, is usually presented as good news for efficiency and bad news for junior designers. But there is a third consequence: if the floor rises and average human attention and aesthetic discrimination decline simultaneously, the practical effect is that the floor becomes the ceiling. Nobody is pushing quality upward because nobody can tell the difference long enough to demand it. A 2026 paper on interpretable multimodal classification (arXiv) notes that affect and behavior classifiers now outperform human evaluators on short-duration aesthetic judgments. The machine is already better at first impressions than you are.
Jenny Odell's Advice Meets the Algorithmic Design Tsunami
Odell's argument in 'How to Do Nothing' was always structural, not motivational. The problem is not willpower. The problem is that every surface of the attention economy is optimized against sustained perception. Now those surfaces are being generated by systems that have been trained on what captures attention, not what rewards it. The arXiv paper on LLM misconceptions by Zhicheng Lin flags this precisely: models embedded in design and content workflows are optimized for engagement proxies, not quality proxies. Kyle Chayka's Culture Slop conversation on algorithmic homogenization called this years ago: taste as deeper understanding cannot survive a feed that rewards only superficial preference. Giving AI 'good enough' taste while humans lose the capacity to demand better is not a design problem. It is an epistemic one.