How do designers develop taste at all if AI removes the repetitions that used to build it?

The shift

Generating plausible visual and interaction options goes from scarce — gated by a designer's hours of manual iteration — to abundant and near-instant. The repetitions that used to be the only route to "what good looks like" now cost nothing, which breaks the mechanism, not the goal.

The axioms

• Taste is built by volume of hand-executed reps, because each rep used to cost real time and effort — scarcity of execution capacity forced slow, cumulative exposure.
• Friction filters ideas: only concepts worth the effort got carried through to completion, so what you learned to judge was pre-selected by cost.
• Juniors build judgment by doing the grunt execution work themselves — the pipeline role and the taste-building role were the same role.
• Taste comes from watching your own choices ship and fail in the real world, and that feedback loop was rate-limited by how much any one person could physically produce.
• Taste is a byproduct of practicing a scarce skill (execution), not a skill anyone trains directly — nobody had to, because the volume constraint did the training for free.

Invalid axioms

1. Taste is built by volume of hand-executed reps. Producing plausible options was the scarce, slow part; AI made it abundant and instant. The habit-trap: studios still structure junior years around "do the reps yourself first," assigning grunt execution as if the doing were what teaches, when the doing is now the cheapest part of the job.
2. Friction filters which ideas survive to be judged. Cost used to kill weak ideas before they reached evaluation, so judgment only ever had to sort a pre-filtered, small set. Now everything survives to the judging stage because nothing costs enough to die early. The habit-trap: reviewing AI output with the same light-touch critique calibrated for a world where reaching the table already meant something had passed a cost filter — it hasn't.
3. The junior execution role is how judgment gets built. Grunt production and taste-formation were bundled into one job because there was no other way to get the exposure. AI unbundles them — you can now generate the exposure without doing the execution. The habit-trap: junior job descriptions and junior pay are still priced against "does the production work," a task that's cheap now, not against the judgment the role was actually supposed to be building.

Unchanged axioms

1. Someone has to decide which of the many plausible options is actually right for this context. AI can generate a hundred defensible directions; it can't tell you which one serves this brand, this user, this business constraint, under this deadline. That call is judgment on novel, situated stakes — still scarce, and now exercised more often per day because more options arrive per day.
2. Feedback from real consequences still has to happen in the world, not in a chat window. A generated option only teaches taste once it's tested against actual users, actual metrics, actual shipped reality — AI can simulate critique but can't replace the physical/transactional loop of something existing in the world and being judged by it.
3. Someone is accountable for the choice. When a design ships and it's wrong, a person owns that, not the model that drafted ten variations. Accountability doesn't get faster or cheaper just because drafting did.

New axioms

1. How do you get calibrated judgment without ever having personally produced the bad version first? The old path (make something mediocre, feel why it's mediocre, improve) assumed you built the thing yourself. If a junior can skip straight to AI-generated competent output, the felt sense of "why this is wrong" — usually earned through your own failed attempts — may never form. Nobody has a replacement mechanism yet.
2. How do you train discernment at the speed options now arrive? When a designer can generate fifty variations in the time it used to take to make one, the bottleneck moves entirely to evaluation — but evaluation skill was assumed to develop from the same slow production cycle that no longer exists. There's no established practice for building a fast, reliable "reject in two seconds" eye at this throughput.
3. What replaces the junior-execution apprenticeship as the entry ramp into the profession? If the traditional first job (grunt production) is now the cheapest task in the pipeline, it's unclear what juniors are for, what they're paid to do, or how they earn the trust and reps that used to come bundled with that job.

Where it breaks

Studios keep hiring and pricing junior designers as production hands (INVALID axiom 3) while the actual open problem is that nobody knows how to manufacture calibrated judgment at all once production stops being the training mechanism (NEW axiom 1). The junior role gets hollowed out — cheap to replace with AI drafts — right as the field most needs a deliberate way to grow the discernment that used to form as a free side effect of that same role.

Separately: reviewers still critique AI-generated first drafts with the light touch calibrated for a world where reaching a review meant surviving a cost filter (INVALID axiom 2), but under new-abundance conditions every draft reaching review is unfiltered by definition (NEW axiom 2) — so the review step is quietly absorbing all the judgment work that used to be distributed across the whole slow production process, and most teams haven't restaffed or resequenced review to carry that weight.

ai-axioms.vercel.app/axioms/product-design/design-junior-taste-development-ai-axioms