Every week, another creative professional sits across from me and tells me the same thing in slightly different words. The art director says it as worry. The editor-in-chief says it as resignation. The brand lead says it as a half-joke. The shape is always the same: AI is about to make my judgment obsolete.
I understand the fear. If a model can write the headline, design the layout, and propose the campaign in fifteen seconds, what is left for the person who used to do all three?
Working with these tools every day and watching brand teams reorganize, I see the value moving toward judgment. Making the first draft used to take eighty percent of the calendar; its cost has fallen to roughly zero. Someone still has to decide which draft fits the audience and deserves to be published. That responsibility becomes more central as production gets cheaper.
When creation becomes abundant, taste becomes infrastructure.
When First Drafts Got Free
The structural fact arrived without a press release. In November 2024, the number of AI-generated articles published on the open web surpassed the number written by humans. Within twelve months of ChatGPT’s launch, AI articles already accounted for 39% of what was being published, on the way to a near 50/50 split. This was an analysis across more than sixty-five thousand articles, not a vendor forecast. It provides one measure of how quickly publishing volume changed.
Basic economics says that when supply of any good rises sharply while demand stays roughly constant, the price falls. The supply of “good enough” creative output — the grammatically correct essay, the on-spec design, the competent campaign deck — has risen sharply.
That abundance increases the demand for judgment: the capacity to decide what is good, what fits the brand, what the audience will actually trust, what deserves to be published into a feed already drowning in adequacy. The supply of opinions about what good looks like has stayed the same. There is simply far more work to judge.
This is the part most coverage misses by treating AI as a productivity story. Productivity gains are the side effect. The real shift is that the scarcity in the entire creative pipeline has moved.
What Is Taste, Precisely?
By taste, I mean the capacity to decide what should exist for a particular brand, audience and moment, and to take responsibility for that decision. This goes beyond aesthetic preference or having “an eye.”
David Droga runs Accenture Song, which makes him the senior creative officer of one of the largest creative organizations on the planet. He argues: “Taste is a real thing. Understanding, context, and emotional connection are still human terrain”. When the person whose business is to scale creative work argues that machines can compress the mediocre middle of formulaic output but cannot occupy the territory of judgment, the claim earns weight.
Jessica Walsh put the operational version of it more bluntly: “AI can generate infinite options. What it can’t do is decide what’s good, what’s wrong, what’s culturally off, or what will age badly”. That last clause concerns the long-term effect of a publishing decision. AI optimizes against the present. Taste accounts for time.
Skeptics will reasonably point out that AI is already learning aesthetic patterns from human-curated examples and may eventually close the gap. The technical objection is real. It also misses what taste structurally requires.
There is an academic version of this too, useful because it explains why audiences punish AI-only output even when the output is technically fine. Researchers studying advertising have named it creative legitimacy — the idea that audience response depends not just on whether AI was used, but on how its participation is “framed, disclosed, and symbolically integrated into the communicative act”. Translated into the language brand teams actually use: people can feel the difference between something a human chose to publish and something a machine produced. They will trust the first and quietly mistrust the second, even when neither is labelled.
The Harvard framing makes the cognitive distinction explicit. Human intuition and taste operate at levels of contextual and embodied experience that current AI systems can approximate through pattern matching but cannot possess. The distinction I am making is that taste is what happens when a person who will be blamed if the call is wrong makes the call anyway.
A model can mimic the pattern of a high-taste editor. It cannot be the person at the meeting on Monday explaining why the campaign was killed.
But What About Distribution?
A counterargument I have heard in these discussions is that taste might matter much less than I am claiming, because distribution and data are doing the work people used to credit to judgment.
It goes like this. In performance channels — paid social, programmatic, growth marketing — A/B-tested mediocrity reliably outperforms editorially curated work. The algorithm rewards what gets clicked, and what gets clicked is not always what an editor would have chosen. If you have superior distribution and superior audience data, you can publish adequate content at scale and win. No taste required.
Distribution and audience data can therefore outweigh editorial judgment in these channels.
My argument applies under a narrower condition: taste is the differentiator among competitors with equivalent distribution and data. Two brands with similar reach and similar feedback loops will be separated by whichever one can tell adequate from excellent — and increasingly by whichever one can resist the gravitational pull of “the metric said yes.”
The New York Times, covering this same shift, captured both halves of it. Excessive AI production produces what one practitioner there called “slop,” and another quoted source put the diagnosis sharper: “Generic taste is inherently bad taste because taste is inherently relational and relative”. Generic taste is what the algorithm trains you toward. It is also, in a market drowning in generic, the worst thing you can have.
“Pretty good is no longer good enough” not because audiences got snobbier, but because the supply of pretty good went vertical. In categories where audiences can tell — and that set keeps expanding — competing on adequacy is competing for the floor.
Organizations Already Restructuring
The New Yorker spent 2024 and 2025 watching its submission volume balloon as AI writing tools proliferated. The publication’s response was not to lower the bar to accommodate the flood. It was to articulate its taste standard more explicitly — what the magazine means by “writing that earns a slot” — and to make that standard institutional rather than implicit. The editorial role, previously a cultural norm, became the value proposition. By all public indications, reader trust held and subscription revenue held — while other publications that loosened editorial standards to chase the volume saw trust erode.
The example suggests that when production costs collapse, the institution that maintains its standards gains differentiation by default. Consistent editorial standards become more distinctive as publishing volume rises.
Accenture Song under David Droga is the corporate-restructuring version of the same insight. As AI handled more first drafts, ideation variants, and versioning, Droga publicly repositioned senior creative roles from maker to tastemaker. Senior people spend less time executing and more time on conceptual origination and final editorial judgment. Efficiency gains from compressed production get redirected into more rigorous taste-based filtering. The business model shifts: the premium charge moves to judgment, not production hours.
The pattern shows up in the workforce data too. Creative professionals heading into 2026 are explicitly investing in editing, curation, and conceptual direction as the skills they expect to keep them employed. These skills emphasize selection and direction over production speed.
There is even some emerging performance evidence, though I would not put my whole case on it. One industry analysis claims that hybrid content workflows — AI drafting, humans editing and selecting — outperform AI-only content by 33% in audience trust and 23% in engagement. These are vendor-reported figures, so I would treat them as provisional evidence for the value of human editing and selection.
Making editorial judgment part of the work
I use four practices with clients to make editorial judgment part of their regular workflow.
Layer 1: Define the Standard. Articulate what “good” means for your specific brand, audience, and context. Not “authentic” or “bold” — operational criteria you can use to resolve disputes. The exercise that surfaces them: take ten recent pieces of work, force-rank them with written justifications, investigate the disagreements rather than averaging them out. The failure mode is aspirational vocabulary that cannot drive a decision.
Layer 2: Install the Filter. Build explicit human decision points into the AI-assisted workflow, with a named accountable person at each gate. Not a final review tacked on at the end — structured judgment at prompt construction, output selection, editing, and approval. The failure mode is checkbox theatre, where reviewers are incentivized for speed and approve everything to clear the queue.
Layer 3: Build Institutional Memory. Log the editorial decisions — published, revised, killed, with one or two sentences of reasoning. After ninety days, the pattern is your taste standard, not the aspirational one in the style guide. The failure mode is key-person dependency, where the senior editor carries the standard in her head and the moment she leaves, the standard goes with her.
Layer 4: Signal Taste Externally. Make the people doing the judging visible to the audience. Named editors, named curators, editorial standards published, correction protocols when you get it wrong. Brand identity is increasingly the primary defensible asset in environments where AI imitates surface style quickly. The failure mode is brand theatre — claiming editorial authority you cannot back up. Audiences in AI-saturated environments are getting unusually good at spotting hollow claims.
An organization may implement one practice or two without giving the people reviewing work a consistent standard. The practices need to support one another: a stated standard, decisions made against it, and a record the next reviewer can learn from.
This framework does not apply to high-volume, low-differentiation content where performance is entirely algorithmic and brand is not a strategic asset. It also does not apply to organizations that have not yet developed a coherent point of view — you cannot formalize a taste standard you do not have. The framework formalizes existing judgment. It does not invent it.
Which Definition Of Yourself Survives?
What is the creative anxiety actually about, underneath the surface fear of being replaced? For a long time, “I am a writer” or “I am a designer” or “I am a creative director” meant I am the person who produces this kind of artifact. That definition is now under serious pressure, because the artifact got cheap to produce.
There has always been a second definition lurking inside the first.
I am the person who decides what should exist.
That definition — the editor inside every maker, the curator inside every creator — is the one that becomes more valuable, not less, as machines take over the making.
Some of this is a fight most organizations will lose. They will not invest in taste infrastructure. They will let AI fill the editorial vacuum with adequate output and ship the result. The taste economy will not be universal. It will be selective.
The companies that build the four-layer stack will outperform the ones that do not. That performance gap will create competitive pressure. The brands whose audiences trust them will pull market share from the brands whose audiences cannot tell whether anyone is home. And the creative professionals who reposition themselves as editors, curators, and tastemakers — rather than competing with AI on production speed — will capture the premium that used to live in the making.
Give the person reviewing the next draft a clear standard, the authority to use it, and examples of earlier decisions. That is how a team can keep exercising judgment when there is far more work to choose from.
References
- Graphite (2025). “More Articles Are Now Created by AI Than Humans.” Graphite
- University of Pennsylvania Wharton (2025). “AI, Taste, and the Future of Creativity: David Droga on What’s Worth Saving.” Wharton AI
- Flint Group (2026). “The Future of Content: AI vs. Human Creativity in 2026.” Flint Group (vendor source — performance metrics cited with caveat)
- Nexus Press / Journal of Language, Communication and Applied Studies (2026). “Human-AI Co-Creativity in Advertising: Authorship, Authenticity, and Creative Legitimacy.” Nexus Press
- Creative Bloq (2026). “These Are the New Skills Creatives Are Learning in 2026 to Save Their Jobs from AI.” Creative Bloq
- Creative Bloq (2026). “These Are the New Skills Creatives Are Learning in 2026 to Save Their Jobs from AI.” Creative Bloq (duplicate of entry 5, present verbatim in the original)
- The New York Times (March 2026). “Is Taste the One Thing A.I. Can’t Replace?” NYT
- Harvard T.H. Chan School of Public Health (2026). “Essay: Intuition and Taste in the Age of AI.” Harvard Chan
- Lateral Action (2025). “AI for Creatives: Why Your Taste Matters More than Ever.” Lateral Action