---
title: "The industry that stopped training its own taste"
locale: "en"
url: "https://irz.fr/en/articles/dad-cost-shortcut-creative-industry-en"
markdown_url: "https://irz.fr/en/articles/dad-cost-shortcut-creative-industry-en.md"
category: "ideas"
tags: ["creative industry", "AI", "apprenticeship", "agencies", "creative economy"]
published_at: "2026-08-24T20:15:00.000Z"
author: "Arthur Lacoste"
translation: "https://irz.fr/fr/articles/dad-cout-raccourci-industrie-creative-fr.md"
---

# The industry that stopped training its own taste

D&AD's 2026 report doesn't say AI is replacing creatives. It documents a quieter mechanism: the tasks being automated were the industry's training program, and nothing has replaced them.

Someone typed the word "ketchup" into a live generative prompt. The model answered Heinz. Almost every time. Forty years of brand-building were enough to make a triangular bottle the default answer to a question asked across billions of parameters.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

The anecdote is buried mid-way through the *AI & Creativity 2026* report that D&AD published on August 18. It sums up what the British design and advertising body spent the year documenting better than the title does: 197 creative leaders — founders, creative directors, strategists, producers, lecturers — interviewed across 30 countries, 82 hours of conversations, cross-checked against an analysis of 10,118 entries to the D&AD Awards.[1](https://www.dandad.org/insights/learning/ai-report-2026)[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/)

> D&AD, 18 August 2026
> **What the report is made of**
> - creative leaders interviewed: 197
> - countries covered: 30
> - of in-depth conversations: 82 h
> - D&AD Award entries analysed: 10,118
> Sources: D&AD, Creative Boom, B&T.

The report is called *The Cost of the Shortcut*. Its core finding is not that AI does bad work. It's that AI does competent, polished, generic work — and that the part of the industry which learned precisely how to tell the difference between competent and good has just been cut from the budget. "AI didn't unlock creativity. It exposed the value of judgement," the authors write in their opening.[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/) The previous edition, built on 291 leaders across 55 countries, ended on a close but sharper line: AI hadn't unlocked creativity, it had exposed who lacked it.[11](https://144450253.fs1.hubspotusercontent-eu1.net/hubfs/144450253/AI%20report/AI%26Creativity_Report.pdf) In one year, the target of the finding shifted from individuals to organisations — which is roughly the whole point of this year's report.

That's a trend-report sentence, so it deserves to be tested. I took it to the primary data: the annual census of UK agencies, the Awards' own entry data, and a Stanford study of American employment, built on payroll data, published six days earlier. All three say things that resemble each other a little too much to be a fad.

## The mean machine

The first finding fits on one line: left alone, a generative model drifts toward the average. The Heinz ketchup result is the shortest possible demonstration — the model doesn't pick Heinz because it's the best ketchup, it picks it because it's the most present one. Forty years of branding settled the question before anyone asked it.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

D&AD's CEO David Patton pushes the idea further, and it's the most unsettling passage in the report: "AI has changed what bad work looks like. It's no longer lazy or derivative in any obvious way – it's competent, polished, and generic, and it's the default output for anyone, anywhere."[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)[4](https://www.campaignlive.com/article/d-ad-report-calls-industry-teach-junior-talent-human-judgment-wake-ai/1967713)

In other words, the visual filter that used to flag weak work no longer works. Bad work now looks like good work. And the question Patton says the report asks isn't "how do we outdo AI" but one step upstream: how do you train someone to tell competent from excellent when competent has become the free, instant output of the entire environment?[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)

Lisa Smith, global chief design officer at London studio Uncommon, compresses the gap into one sentence: "There seems to be far more conversation about AI than there is genuinely great work being made with it."[1](https://www.dandad.org/insights/learning/ai-report-2026)

The Awards data backs the suspicion, and we'll come back to it: AI is everywhere in the entries, far less decisive among the winners.

## The work was the training

The heart of the report is a section plainly titled "The work was the training". It deserves quoting at length, because it names a mechanism the industry has rarely been willing to call training:

> "The tasks being replaced built something. Layouts, first drafts, mockups, research and iteration build the instincts that matter later. Repeating that work over time is how juniors become seniors. The repetition was the curriculum."[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

For decades, the creative industry treated learning as a by-product: the junior produced the adaptations, the retouches, the reference hunts, and judgement was assumed to arrive by osmosis. The report says it plainly: the osmosis was the curriculum. Not a side effect of cheap labour — its main function from the industry's point of view.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

> Illustration: Composing room of the New York Times in 1942, with several Linotype operators at work. The New York Times composing room, 1942. Material trades never believed repetition was wasted time. Credit: [Marjory Collins, Library of Congress — public domain](https://commons.wikimedia.org/wiki/File:Composing_room_of_the_New_York_Times_newspaper.8d22721v.jpg).

Anyone coming from a material trade won't learn anything new here. In a print shop, a tattoo studio or a pottery workshop, nobody ever believed the years of repetitive gestures before the flagship piece were administrative delay. Repetition is when hand and eye calibrate, when you learn what "too much" means on a specific material. The D&AD report simply does for advertising and design what the craft trades have always known: removing repetition doesn't remove a chore, it removes a manufacturing step — except the thing being manufactured here is the professional's eye.

A professor quoted in the report states the consequence without detour: "AI is lessening opportunities for junior creatives to enter the field. In a few years, when the current juniors and mid-level creatives move up, no one is replacing them. So the industry seems to be eating itself from within."[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

## The base erodes

That diagnosis could have stayed a collection of creative directors' impressions. It happens to have numbers, and they come from a source D&AD doesn't control.

The IPA's annual census — the UK Institute of Practitioners in Advertising, 66th edition, data as of 1 September 2025 — gives this snapshot: total headcount at member agencies fell from 26,787 to 24,963, a 6.8% drop in a year. The fall is much steeper among employees aged 25 and under: from 3,632 to 2,936, down 19.2% in one year. That cohort now makes up 11.8% of the sector's workforce, against 17.4% three years earlier. And only 43.4% of agencies reported still employing graduate trainees, apprentices or school-leaver hires in 2025, down from 56% a year earlier.[6](https://ipa.co.uk/knowledge/publications-reports/agency-census-2025)[7](https://ipa.co.uk/news/agency-census-2025/)

> IPA Agency Census 2025
> **The pyramid, measured**
> - under-25 headcount in one year (3,632 → 2,936): -19.2%
> - share of under-25s, against 17.4% three years earlier: 11.8%
> - agencies still employing apprentices or trainees (56% in 2024): 43.4%
> - creative non-media agency headcount in one year: -14.3%
> Source: IPA Agency Census 2025, data as of 1 September 2025.

The D&AD report adds two American figures: the share of advertising and PR jobs held by 20-to-24-year-olds fell from 10.5% in 2019 to 6.5% in 2024, and more than half of agencies have slowed or frozen entry-level hiring altogether.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

One honesty check is required here, and the report performs it itself: only a minority of agencies blame AI directly. In the UK, just 8% cite AI as the direct cause of headcount reductions; industry consolidation, holding-group cost-cutting and margin pressure are major factors too.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/) 2025 also saw the Omnicom acquisition of IPG completed, followed by thousands of job cuts — a purely financial mechanism that needs no generative model.[8](https://uk.themedialeader.com/agency-employment-declines-6-8-as-creative-roles-hollowed-out/)

But this is where the report's reasoning gets interesting. The question isn't what percentage of the decline AI causes directly. It's that the tasks through which a junior used to learn — adaptations, first drafts, research, iteration — are leaving the junior job description, whatever the reason the job itself is shrinking. "While entry-level jobs are disappearing, AI is simultaneously absorbing the remaining repetitive tasks. Together, this is eliminating the training ground that builds the judgement required for future creative leaders," the authors write.[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)

Two independent causes reinforcing each other, in short. The pyramid hollows from the base, and what little is left at the bottom trains nobody.

## The melting pencil

What remains is explaining why companies that claim to invest in talent cut exactly there. The report's answer fits in one number: 70% of agencies haven't changed what they charge, even though AI has made the work faster.[1](https://www.dandad.org/insights/learning/ai-report-2026)[4](https://www.campaignlive.com/article/d-ad-report-calls-industry-teach-junior-talent-human-judgment-wake-ai/1967713)

> Illustration: Giant pencil sculpted from cracked melting ice on a film set, next to a slippery when wet sign. The D&AD pencil in ice, the report's visual for the value section. The value melts faster than the price moves. Credit: [D&AD, AI & Creativity Report 2026](https://www.dandad.org/insights/learning/ai-report-2026).

The mechanics are simple, and worth unpacking because they connect everything else. If the work gets twice as fast and the price doesn't move, the efficiency gain doesn't stay in the agency: it goes to the client, who gets the same delivery faster for the same price. The report states it as-is: the efficiency gain has been transferred to clients.[1](https://www.dandad.org/insights/learning/ai-report-2026) Meanwhile fixed costs — senior salaries, rent, tooling — don't fall. The only cost line an agency can cut without touching its immediate production capacity is the bottom of the pyramid: the people being trained, precisely.

The report observes that the agencies doing best are doing the opposite: they charge for the decision and the IP, not the delivery.[1](https://www.dandad.org/insights/learning/ai-report-2026) That is, they stop selling hours of production — now nearly free — and sell what remains scarce: the choice, the taste, the accountability for the outcome. Ben Cooper, the report's author, connects that economy explicitly to the training problem: "AI is brilliant at removing friction from creative work, and friction is where judgement is built. Take out the slow, unglamorous parts and you also take out the place where people learned the difference between an answer and an idea."[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/)

There's an irony the report doesn't spell out quite so bluntly, so I'll offer it as analysis: an industry billing by the hour cannot afford to employ people who are learning, because learning is paid for in hours. An industry billing for judgement has no choice but to keep training it, because judgement is its only stock. The pricing model and the talent pipeline aren't two topics. They're one topic.

## What the Pencils say

The report's third data source is the most interesting one for an IRZ reader, because it comes from the competition itself: this year, D&AD analysed the entries to its own Awards, presented as the global benchmark for design and advertising excellence.

Entries declaring AI use more than doubled in a year: from 10.7% to 27.6% of the total.[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/) But the number that matters sits elsewhere. Among AI-declaring entries, 56.2% used generative AI. Among winning entries — the Pencils — that share drops to 44.7%, with assistive AI rising proportionally among winners.[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)

> D&AD Awards 2026
> **Entering and winning don't use the same AI**
> - entries declaring AI, against 10.7% a year earlier: 27.6%
> - generative AI share among AI-declaring entrants: 56.2%
> - generative share among AI-declaring winners: 44.7%
> - AI rising among laureates, alongside craft: Assistive
> Sources: D&AD Awards data relayed by Creative Bloq and B&T.

The most cautious reading: raw generation doesn't win a craft competition. The more interesting one: the people doing award-level work use AI as a tool under their hand — retouching, iterating, exploring — rather than as the source of the work. The report draws a measured conclusion: "its use alone does not translate into creative excellence"; AI is "neither a shortcut to excellence nor a prerequisite" for it.[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)

Which is exactly what the opening line said — AI exposed the value of judgement — except with data behind it. The work that wins is increasingly work where a person decided, selected, corrected. And the judgement that does the deciding, first section obliges, gets built in the very jobs being cut.

## The same crack

What's happening in British advertising could have stayed a local story. Except that six days earlier, Stanford's Digital Economy Lab had released the annual revision of a study measuring the same phenomenon across the entire US economy, using payroll data from ADP — millions of workers.[9](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/)[10](https://digitaleconomy.stanford.edu/news/canariesaug26/)

Their finding: employment of 22-to-25-year-olds in the most AI-exposed occupations now sits 19% below where it would have been had it kept pace with less-exposed occupations. A year earlier the gap was 15%: it's widening, not closing. And the detail that counts: the adjustment runs through reduced hiring, not layoffs. Nobody is being fired in favour of a model; people are simply never hired.[9](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/)[10](https://digitaleconomy.stanford.edu/news/canariesaug26/)

> Stanford Digital Economy Lab, August 2026
> **The same hole, whole economy**
> - employment of 22-25s in exposed jobs, below the kept-pace counterfactual: -19 %
> - the gap widened in one year: 15% → 19%
> - the adjustment channel, not layoffs: Hiring
> - declines concentrate where AI substitutes, not complements: Automation
> Source: Brynjolfsson, Chandar & Chen, ADP data through June 2026.

The study goes one step further into mechanism, and that's where it meets D&AD without having read it: the declines concentrate in uses where AI substitutes for human work, not where it complements it. And the authors offer a hypothesis worth taking seriously: generative AI substitutes effectively for codified knowledge — the kind learned from manuals, procedures, standardised tasks — and complements tacit knowledge — the kind acquired through repeated practice, mentorship, exposure to real situations.[9](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/)

Translated into D&AD's vocabulary: codified knowledge is everything a junior learned by producing; tacit knowledge is the judgement the report wants to save. Two studies built independently — one on interviews with creative leaders, the other on payroll data — describe the same asymmetry. The bottom of the skill ladder is precisely the part the models can already do, and it was also the part you entered the trade through.

## Who pays for this?

Before going further, three warnings, because a report like this never arrives without interests.

First, it is sponsored by Shutterstock Studios, a supplier of creative tools with a precise interest in the industry adopting AI "properly" rather than rejecting it. The company's VP, Adam Barnett, is quoted in the report: "We need to use AI to remove unnecessary friction, but protect the parts of the process where creative judgment is built."[1](https://www.dandad.org/insights/learning/ai-report-2026) A tool vendor recommending moderate use of its own tools: the position is understandable, it isn't neutral.

Second, D&AD is an educational charity selling training and programmes for young creatives. A report on a training crisis is also, mechanically, a shop window for that offer.

Third, causality isn't proven, and the Stanford authors are the first to say so: their figures are "descriptive indicators", not causal estimates, and they note divergent trends predating ChatGPT themselves.[9](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/) For British advertising, sector consolidation explains a good share of the headcount fall across all cohorts.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)[8](https://uk.themedialeader.com/agency-employment-declines-6-8-as-creative-roles-hollowed-out/)

With those reservations on the table, the picture still holds, for one simple reason: the sources don't copy from each other. The IPA census is an annual administrative dataset that predated the report.[6](https://ipa.co.uk/knowledge/publications-reports/agency-census-2025) The Stanford study works on payroll data with no link to the creative industry.[9](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/) And the Awards data is internal to the competition. Three independent measurement devices, three methods, one convergence. That's rarely the sign of a fad.

## Redesign the ground floor

The report's proposal is not to protect junior jobs as they were. It's more uncomfortable: "The fix isn't to protect those jobs as they are – it's to redesign them, so they still teach the judgment the industry depends on."[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)

Concretely, the report suggests rebuilding junior roles around critique, taste and cultural awareness — the skills automation doesn't supply — rather than defining them by the production tasks it has absorbed.[3](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)[4](https://www.campaignlive.com/article/d-ad-report-calls-industry-teach-junior-talent-human-judgment-wake-ai/1967713) The sharpest commentary on the report, published in Australia by B&T, offers a test that works for any practice, not just advertising: audit your tasks one by one and ask which were waste and which were formative. Automate the first, deliberately reinvent the second.[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/)

The report also insists on a point that sounds soft and is in fact structural: leaders are deploying tools they don't master themselves, to teams who see that uncertainty perfectly clearly.[1](https://www.dandad.org/insights/learning/ai-report-2026)[5](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/) B&T draws the consequence: a creative who thinks a mistake will end up as evidence against them stops showing drafts, stops trying the strange routes, and optimises the appearance of competence. But apparent competence is exactly what AI already produces for free. Psychological safety isn't an HR comfort here: it's the condition for someone accepting to produce a bad first draft in front of witnesses — and the bad first draft, the whole report argues, is the raw material judgement is made of.

Patton sums up, in the report's presentation, with a line that has the merit of posing the problem without solving it: "AI has raised the floor. It's up to all of us to decide where the ceiling should be."[1](https://www.dandad.org/insights/learning/ai-report-2026)

## The 2031 bill

What convinced me this report deserved more than a signal wasn't its diagnosis — the idea that AI averages work out is already well established. It's the silent arithmetic behind it.

An industry hiring 19% fewer juniors in a given year doesn't have a recruitment problem. It has a dated problem: the people who should have entered in 2025 will be the project leads, art directors and reference craftspeople of the 2040s. The report predicts that gap is already locked in.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/) Creative Boom summarises it with a sentence D&AD couldn't have disowned: the juniors nobody hired in 2025 are the seniors who won't exist in 2031.[2](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)

You're allowed to dislike ad agencies. The mechanism doesn't stop at agencies: the Stanford data finds it in software development, customer support and other exposed desk jobs — everywhere entry into the trade ran through tasks the models can now do.[9](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/) And for anyone practising a material trade, the report tells a familiar story in new vocabulary: a skill doesn't transfer by downloading a tool. It transfers by letting someone do the thing, badly, for a long time — the gestures that look unproductive, until the day they turn out to be the ones deciding whether the work is good.

The report's open question isn't technological, then. It fits in one sentence: if everyone wants to buy judgement and nobody wants to train it, who, exactly, is going to produce it?

## References

1. [D&AD, AI & Creativity Report 2026: The Cost of the Shortcut, 18 August 2026](https://www.dandad.org/insights/learning/ai-report-2026)
2. [Creative Boom (Tom May), D&AD has a stark message for creatives: the industry is eating itself from within, 18 August 2026](https://www.creativeboom.com/news/dad-has-a-stark-message-for-creatives-the-industry-is-eating-itself-from-within/)
3. [Creative Bloq (Joe Foley), Replacing creative jobs with AI could have a hidden cost, a new report warns, 19 August 2026](https://www.creativebloq.com/ai/replacing-creative-jobs-with-ai-could-have-a-hidden-cost-new-report-warns)
4. [Campaign (Charlotte Rawlings), D&AD report calls on industry to teach junior talent 'human judgment' in wake of AI, 18 August 2026](https://www.campaignlive.com/article/d-ad-report-calls-industry-teach-junior-talent-human-judgment-wake-ai/1967713)
5. [B&T, Automating The Apprenticeship & Assuming Mastery Arrives Anyway, 18 August 2026](https://www.bandt.com.au/automating-the-apprenticeship-assuming-mastery-arrives-anyway/)
6. [IPA, Agency Census 2025](https://ipa.co.uk/knowledge/publications-reports/agency-census-2025)
7. [IPA, IPA Agency Census 2025 shows workforce declines while diversity improves](https://ipa.co.uk/news/agency-census-2025/)
8. [The Media Leader (Jack Benjamin), Agency employment declines 6.8% as creative roles hollowed out, 11 February 2026](https://uk.themedialeader.com/agency-employment-declines-6-8-as-creative-roles-hollowed-out/)
9. [Erik Brynjolfsson, Bharat Chandar & Ruyu Chen, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, Stanford Digital Economy Lab, revised August 2026](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/)
10. [Stanford Digital Economy Lab, No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%, 12 August 2026](https://digitaleconomy.stanford.edu/news/canariesaug26/)
11. [D&AD, The AI & Creativity Report 2025 (PDF)](https://144450253.fs1.hubspotusercontent-eu1.net/hubfs/144450253/AI%20report/AI%26Creativity_Report.pdf)