Will AI automation eventually formalize designer taste?

Designers under pressure from AI automation often retreat to taste as the protected core of their work. AI can produce layouts, generate variations, draft copy, even prototype interactions — but it cannot have taste. Taste is judgment shaped by lived experience, by years of seeing what works and what does not, by an intuition no training distribution can produce. Therefore (the argument runs) the human designer remains essential as the holder of taste even as execution is automated.

The argument inherits a structural pattern from prior automation waves and predicts the same outcome. Each wave begins with practitioners identifying a core capacity that "machines cannot replicate." Each wave proceeds by formalizing that capacity into a process that machines can then execute. The capacity is not preserved as the protected core; it is converted into the next layer of work to be automated.

For taste specifically, the formalization mechanism is already visible: evals. AI workflows increasingly require evaluation infrastructure to assess output quality. Designers who want to retain influence over AI design output will do so by writing the evals — encoding their taste into criteria, examples, rubrics, and preference data that the AI uses to score and improve its output. The taste does not stay in the designer; it migrates into the eval, and the eval is then applied automatically. The reification is what allows the taste to scale; the scaling is what makes the human-with-taste optional.

This is not a critique of designers writing evals — they should, because writing evals is how their judgment continues to shape outcomes. It is a correction to the argument that taste is the protected core. Taste is the next thing to be formalized, not the thing that resists formalization. The designers who survive the next wave are not the ones whose taste cannot be encoded; they are the ones whose taste is the most worth encoding, who get to write the evals everyone else's AI uses. The position changes from in-the-loop executor to out-of-the-loop authority — fewer roles, more leverage, and a different kind of work.

The diagnostic implication: claims that "AI cannot have X, therefore X is the protected human role" should be read as descriptions of the next layer of automation rather than as defenses of human work. The taste argument is structurally identical to "AI cannot drive cars" or "AI cannot diagnose disease" arguments from prior decades, and will likely follow the same trajectory.

The strongest counterargument: some forms of taste are too contextual to formalize — the designer's judgment about a specific client, a specific brief, a specific moment. True for the most contextual work, but most professional-design work is calibrated to genres and conventions whose taste is precisely the kind that formalizes. The formalization does not need to capture all taste; it needs to capture enough taste to handle the high-volume cases. The contextual remainder shrinks as the formalization improves.

Source: Reification of taste