Taste Labs goes viral trying to end AI slop — hosts say it'll print in the short term despite the backlash

Taste Labs Goes Viral on a Problem That Can't Be Solved—But the Business Model Works Anyway

A former Eksa AI Labs founding team member launched Taste Labs with a straightforward mission: end AI slop by turning aesthetic judgment from subjective intuition into measurable, codifiable data. The startup is working with frontier AI labs through design-focused data labeling and partnering with app-layer startups to improve the visual output of their products.

The post went viral immediately, generating over a million views on X in 24 hours and igniting a familiar debate: you cannot program taste, critics argue, because it is inherently subjective and emerges from craft and genuine care. The counterargument is simpler: improving AI's aesthetic output is valuable regardless of whether you call it "taste" or something else, and startups can absolutely build revenue around that problem.

The real tension is that the word "taste" itself has become fatiguing in tech discourse. For the past year, it has functioned as a catch-all response to the question of what separates human judgment from machine capability. Outside Silicon Valley, the irony cuts deeper—the same people who cite "taste" as AI's missing ingredient are perceived as working in an industry optimized for efficiency over aesthetics, defined by t-shirts and athleisure rather than any recognizable design sensibility.

The commodification trap. Taste, once established, collapses into copy. Linear set a standard for design-driven product development. Then an entire generation of companies simply replicated Linear's aesthetic, making the original taste indistinguishable from the derivative. The same pattern played out with Squarespace, which democratized high-end web design and immediately made every Squarespace site identifiable as such—useful, but no longer tasteful in any meaningful sense. The moment taste becomes reproducible at scale, it ceases to be taste.

Yet Taste Labs will likely print in the short term regardless. Users care about how AI-generated outputs look. Frontier labs care about improving their models' visual quality. Hyperscalers care about the problem. Viral backlash does not suppress business opportunity—it tends to explode pipelines. Whether the company survives long enough to matter in five years is an open question, but the near-term commercial case is sound.

The data labeling approach itself is not new. Previous projects have encoded functional questions: does the button work, does it render properly, does the photo have the right number of fingers. Taste Labs is attempting to scale that into subjective territory—identifying what looks good and, harder still, ensuring AI models can generate visual diversity without collapsing into a single dominant aesthetic style. That last point is the real risk: everything AI touches visually becomes clockable as AI-generated, which means the training approach itself may be fighting against a fundamental problem that no amount of labeling solves.