The Taste Gap.
Exploring how AI's collapse of production costs has flipped the scarce resource from output to discernment, and why the environments that used to build taste are the ones being automated away first.
The workslop economy
In September 2025, Harvard Business Review published a number that should have terrified every knowledge-work organisation in the western economy: 40% of desk workers had received AI-generated “workslop” in the previous month. Content that looked polished but lacked substance. Decks padded to look complete. Memos with the shape of analysis but no spine. Reports where the conclusions had been generated before the evidence.
The average worker spent 3.4 hours per month cleaning it up — triangulating sources, re-running numbers, rewriting sections that only sounded finished. For a 10,000-person company, HBR calculated the cost at $8.1 million per year. Two months later, Merriam-Webster quietly marked the moment by naming “slop” its 2025 word of the year: “low-quality AI-generated content flooding online spaces”.
But the dollar figure understates the problem. The cost isn’t the hours. It’s what those hours required.
To clean up workslop, you need taste.
You need the ability to look at a shiny-looking output and feel — in the prose, in the structure, in the argument — where it has gone wrong. You need a calibrated sense of what a good memo reads like, what a coherent deck argues, what a deliverable that earns trust actually contains.
And here is the trouble: most organisations are operating with less taste than they had five years ago. Not more.
This is the most important thing happening in knowledge work right now, and it barely has a name.
Let’s call it the Taste Gap.
The abundance flip
For every generation of knowledge workers before this one, the scarce resource was production. Could you write this quickly enough, with enough polish, at the required volume? Could you design it, code it, model it, illustrate it? Time, skill, and raw output capacity were the bottleneck.
That bottleneck is gone.
In 2026, a moderately-skilled practitioner with the right tools can generate, in a single afternoon, what used to be a fortnight of work from a well-staffed team. Decks, research briefs, first-draft strategies, landing pages, prototype code, visual identities, internal memos, data summaries — all close to free. Not perfect, but close enough that the variance between “good” and “mediocre” is no longer bridged by doing the work. It’s bridged by knowing what good looks like.
This is what I’ve come to call the abundance flip. When you can generate anything, the only remaining question is what’s worth committing to. And that’s a taste question. Not a production question.
The designer and writer at Designative put the shift crisply:
“Taste is the judgement that operates when options are abundant — when many solutions are technically viable, data-backed, and defensible. It’s what allows teams to discriminate between them, to explain why one direction deserves commitment while others do not.”
For a workforce trained primarily to produce, this is an unexpected pivot. We optimised for output for a century. Taste was the thing you picked up informally — from the partner who’d return your draft covered in red ink, from the VP who’d kill your concept and tell you why, from long hours in review comparing five takes to pick the one. Taste was a side-effect of production. An apprenticeship dividend. It was never the job itself.
Now it’s the job. And most of us are underqualified.
What taste actually is
Before we go further, it’s worth being precise, because “taste” is a word that has absorbed too much mystification.
Taste is not a vibe. It isn’t subjective. It isn’t “knowing what you like.”
Taste a learnt sensitivity to context, audience, and consequence, developed through prolonged exposure, critique, and revision.
Nielsen Norman Group calls it a decision-making skill. Anders Ericsson, the psychologist who founded the field of deliberate practice, would have recognised it as the output of thousands of cycles: attempt, feedback, reflection, refinement. The research on expert performance is clear: experts aren’t born with taste; they’re built through mentored repetition in high-feedback environments.
You can split taste into four working varieties, each at a different stage of decay:
Contextual taste — knowing what’s right for this audience, this organisation, this moment. The instinct to recognise that a deck that would slay in Amsterdam will die in Tokyo; that the Friday-afternoon email wants a different register from the Monday-morning one.
Editorial taste — structural judgement. Knowing what to cut, what to emphasise, what to reorder. Feeling when an argument has a hollow middle, or when the second paragraph is doing the work the first should have done.
Aesthetic taste — sensory judgement. Knowing what reads right, what sounds right, what looks right. Not “pretty” but calibrated. The reason two versions of the same deck provoke different reactions even when the content is identical.
Strategic taste — discernment about what’s worth doing in the first place. Which problems are actual problems. Which questions are worth asking. Which bets are worth making. This is the highest-stakes form of taste, and the one AI has least access to, because it’s fundamentally a question of what matters, and AI has no stake in what matters.
All four are degrading as we outsource the practice that built them.
The apprenticeship vacuum
Here’s the most uncomfortable part.
Ira Glass — the radio producer — famously articulated what he called “the gap” for creative beginners: people enter a field because their taste is already sophisticated. They can tell good work from bad. Their problem is that their output doesn’t yet match their taste. That’s the gap.
His advice was the only advice that has ever worked: do a huge volume of work. Put yourself on deadlines. Accept the discomfort of producing things you know aren’t yet good. Eventually, your output catches up to your taste.
Two decades later, we are watching an inversion of that problem unfold in real time. AI is doing the production. Beginners don’t have to sit in the gap any more. They don’t have to push through the discomfort. They don’t have to produce ten bad decks in order to internalise, viscerally, what a bad deck is and why.
This sounds like progress. It is a catastrophe for taste formation.
Taste does not form by consumption alone. You don’t get it by reading great work; you get it by trying to make great work, failing, comparing your output to the best of the field, and feeling — physically, uncomfortably — where your attempt fell short. You get it through a ten-year cycle of production-feedback-revision-production. The work itself is the training set.
And the work itself is exactly what we are liquidating:
The junior analyst who used to spend eighteen months pattern-matching across hundreds of client decks? AI drafts the deck now. She never sees the hundred decks.
The associate designer who used to generate fifty variations of every logo mark? AI does it in thirty seconds. He never develops a feel for the shape of what works.
The editorial assistant who used to read two thousand submissions to find forty good ones? AI pre-filters. She never builds the eye.
The new partner who used to sit in every pitch meeting, absorbing how senior partners chose and cut and defended? Those meetings are now abbreviated or auto-summarised. He never sees the cuts that mattered.
We’ve eliminated the apprenticeship without naming what we’ve eliminated.
The production work was never just production. It was the scaffolding on which taste was built. Remove the scaffolding and you don’t get taste more quickly. You get taste not at all.
Call this the Apprenticeship Vacuum. It is the defining risk of the next decade of knowledge work, and almost no one is managing for it.
The calibration crisis
A second, quieter problem runs parallel to the first: we are losing our sense of what “good” even means.
When every deck looks competent, competence loses its signal. When every email reads polished, polish becomes noise. The reference points that knowledge workers once used to calibrate their own standard — that deck from a senior partner, that memo from the CEO, that essay you remembered a decade later — are drowning in a sea of adequately-produced everything.
This is what the “AI slop” discourse is really about. It’s not that AI output is uniformly terrible. Most of it is mediocre-to-decent. The problem is that mediocre-to-decent is now the ambient baseline. Our sense of “great” is eroding because we can no longer easily find the edge cases that used to anchor it. The peaks look lower because the valleys have risen.
Europol has projected that by the end of 2026, as much as 90% of online content may be synthetically generated. Even if you discount that number significantly, the directional truth holds: we are about to live in a world where most of what we read, see, and evaluate at work was produced by systems with no stake in any of it. Calibration under those conditions is not automatic. It requires effort.
Organisations used to run on implicit calibration. Reviews, edits, critiques — these transmitted, week by week, what the house standard was. When that process is automated or abbreviated — “the AI can redraft it” — the calibration stops happening. Teams drift. Standards don’t fall all at once. They fall one unreviewed deliverable at a time, for years, until one day a senior leader opens a deck and doesn’t understand why it feels so hollow, even though every box is ticked. By then, the people who would have told them why are five years gone.
A counter-argument, honestly considered
“Every new tool triggered this panic,” the sensible person says. “Photography was supposed to kill painting. Calculators were supposed to kill arithmetic. Spell-check was supposed to kill spelling. None of it happened. People adapted. Taste migrated. Why should this be different?”
It’s a fair challenge and worth answering directly.
The honest answer is: the earlier tools removed discrete, bounded capacities. A calculator does long division. A spell-check checks a word. Each replaced one small layer of cognitive work, leaving the surrounding judgement largely intact. You still had to decide which equation to set up, which sentence to write, which argument to make.
Generative AI is different in kind, not degree. It removes the whole surface between initial intent and finished artifact — including most of the middle-skill judgement calls where taste is forged. A junior designer using Photoshop in 2010 made hundreds of micro-choices per day: font weights, kerning, colour relationships, negative space, hierarchy. A junior “designer” using a generative tool in 2026 may make a handful of prompt-level choices and pick from four options. The volume of calibration reps per day has collapsed by an order of magnitude — and it’s the reps, not the output, that built the designer.
That is what makes this particular substitution dangerous in a way that calculators never were. We are not removing a tool. We are removing a gym.
The discernment dividend
There is, however, a bright side hidden inside this — and the organisations that find it first will own the next decade.
Taste is getting scarcer, and scarcity prices value. The Discernment Dividend is the compounding economic premium accruing to people and organisations with calibrated judgement in a world where everyone else can produce but fewer can discriminate.
Signs of it are already visible. Editors are being paid more, not less, in AI-saturated publishing. Senior designers command higher multiples over juniors than they did in 2022. “Curator” roles — people whose sole job is to choose and defend — are appearing in product, publishing, and learning organisations. The creator economy is quietly bifurcating between high-volume generators (low margin, low defensibility) and taste-driven brands (high margin, fiercely defensible).
This is the Discernment Dividend starting to show up in pay packets. It will accelerate.
Practical implications
For early career: your production ability no longer differentiates you. Your taste does. Treat taste-building as the core of your first decade, not a by-product of it.
Consume excellent work constantly — not passively, but analytically. Why is this piece good? What decisions did the writer make? Where would a worse version have drifted? Keep a private file of work that moved you, and revisit it. Make notes on what specifically landed.
Seek critics. Find the person in your organisation whose taste you most respect and ask them to shred one piece of your work every month. Do the work AI can’t yet: original hypotheses, unexpected framings, critique that takes a risk.
And do some work by hand, sometimes. You will be slower. You will be right less often. You will learn what it feels like to struggle with a problem — which is the only way taste gets installed.
For mid-career: you are at the most dangerous inflection of your career. Your taste is partially built. Your role is being restructured to lean harder on AI. You will be tempted to coast on the taste you already have while AI handles the execution.
Don’t.
Taste is a muscle. It atrophies. The professionals who will matter in 2035 are not the ones who optimised for AI-assisted output in 2026. They are the ones who kept showing up to the work where taste is tested — live critiques, genuine disagreements, decisions under real stakes. Resist the drift toward being a “reviewer of AI drafts.” You will degrade into it if you’re not careful.
For hiring: stop screening for production skills. Everyone’s writing samples look good now. Everyone’s portfolio is polished. Screen for discernment. Show candidates three pieces of AI-generated work and ask them to rank and defend. Present a flawed strategy and ask what they’d cut and why. The person who can articulate why one version is better — and can do it in a way that changes how you see the work — is worth five who cannot.
Interview for critique, not composition.
For leaders: you are running a taste-development programme whether you named it that or not. Every review is a training signal. Every “ship it” teaches your team what good means to you. If you outsource your reviews to AI summaries, you have stopped teaching taste in your organisation. Full stop.
Consider actively protecting apprenticeship work. Keep some decks hand-drafted. Keep some critiques human. Make exposure to your best people’s reasoning a formal benefit of working at your company, not an accident. The companies that do this will quietly collect the strongest talent — because good people want to get better, and they can only get better somewhere that still teaches taste.
For organisations: audit your AI investment. For every dollar you spend on production tools, how much are you spending on taste development — on critiques, on reviews, on exposure to excellent work, on the meetings and moments that transmit standards? If the ratio is 100:1 in favour of production, you are over-indexed on the thing that has become commodity and under-indexed on the thing that has become moat.
Name “taste” as a strategic capability. Measure it — not with vanity metrics, but with what your best reviewers say about the quality of the work shipping across the org, month over month. Appoint senior people to its cultivation. Build it into hiring, promotion, and performance review. The same rigor you bring to AI adoption, bring to discernment cultivation.
And consider protecting the humble, unglamorous rituals that actually build taste: the weekly deck review where someone says “this section is wrong and here’s why”; the portfolio critique; the editor who line-edits a draft in front of its author; the post-mortem where “what did we almost ship?” is asked as seriously as “what did we ship?” These rituals look like overhead on an efficiency dashboard. They are the only reason your organisation will have taste ten years from now.
Most organisations in 2026 are investing heavily in AI tools to increase production. Almost none are investing, deliberately and at scale, in taste.
That is exactly backwards.
Production is the new commodity. Taste is the new moat.
And unlike AI capability — which compounds in weeks — taste compounds slowly, across years of deliberate practice in environments that reward judgement. By the time you realise you need it, it’s a decade too late to build.
We are living through a once-in-a-generation inversion of what’s scarce. The organisations that recognise it will get quieter about productivity gains and louder about standards. They’ll pay more for discernment than for output. They’ll protect apprenticeship even when it looks inefficient. They’ll treat every senior-junior review as strategically important, because it is.
The organisations that miss it will generate more than ever and land less. They’ll wonder why their output feels hollow, why their best people keep leaving, why the work doesn’t cut through anymore. They’ll blame the market, the economy, the competition.
The real answer will be simpler and harder.
They lost their taste. And they did it in a way that felt, every single quarter, like they were winning — more output, more decks, more campaigns, more content shipped per headcount than ever before. Which is exactly why almost no one will notice until the damage is too compounded to reverse.
The window to act is short.
Taste that’s already built can still be deepened. Taste that isn’t yet built can still — for another few years — be installed through apprenticeship, if we choose to protect it.
Past that, we are rearing a generation of knowledge workers who have never once had to stare at a bad draft of their own work and feel what it meant. And no amount of AI will teach them what we decided, through efficiency, to stop teaching ourselves.