The Middle Is Dying. What Remains.

Lucy Jameson, founder of Uncommon, described it this way in a 2025 interview with The Drum: “You’re going to see a sort of bifurcation in the industry.” A bifurcation, a split into two opposing poles. Her prediction was explicit: “There won’t be these people in the middle. There’ll be the people just using tech, data, AI and running everything through a machine.” And to the agencies that refuse to choose: “We have to be really good at what we do. Otherwise, we’re going to get eaten alive by AI.”

By now, most people understand what she means. Few are willing to follow the argument through to its conclusion.

Forrester has put numbers to the shift: an 8 percent reduction in agency jobs in 2025, with 15 percent forecast for 2026. The three largest holding companies are reinventing themselves as platform providers, moving away from cultural production and towards infrastructure services. The bifurcation that many still treat as a future scenario is already under way. At one end are agencies built on distinctiveness, judgement and emotional resonance. At the other is AI-powered production: fast, cheap and scalable. The middle, solid, dependable and lacking a clear point of view, is disappearing.

What this means in practice was on display in Cannes at the end of June. While the year’s most creative campaigns were winning Lions inside the Palais, Meta was presenting a fully automated advertising system on the other side of the Croisette. Its AI combines product data and creative assets in real time to produce the highest-performing ad variation. At the table was WPP, once the world’s largest agency group, not as an opponent, but as a partner. “AI is not a threat. It’s a tool,” says WPP executive Elav Horwitz, sounding rather like AI herself. The agency’s role is shifting from producer to orchestrator.

But that describes what is happening, not why.

Philosopher Dorothea Winter researches the intersection of AI, creativity and ethics. For episode 194 of What’s Next, Agencies?, she asked the decisive question: What can AI not do?

According to researcher Margaret Boden’s model, there are three forms of creativity:

• Combinational creativity: bringing existing elements together in new ways.
• Exploratory creativity: pushing an existing system to its limits and discovering something new in the process.
• Transformational creativity: creating something genuinely new that simply did not exist before, a new genre of music, a new business model or an idea that redefines an entire category.

AI learns from what already exists. It can recognise patterns, generate variations and refine them. That is combinational creativity, and AI does it better than almost any human. It shows signs of exploratory creativity too. One well-known example is AlphaGo’s Move 37 in its 2016 match, a move no human expert would have played because it defied every established heuristic. AlphaGo won the game regardless, or perhaps precisely because of it.

But AI cannot break rules it does not know. Transformational creativity, entering territory for which no map yet exists, remains human. That is not a romantic defence mechanism. It is a philosophically grounded proposition.

Another research finding shows what this means in practice. In an experiment involving short-story writing, people with lower levels of creativity benefited most from AI support. Their stories improved noticeably. The most creative participants benefited the least. AI raises the floor. The ceiling stays where it is, while the middle gets crowded.

Agencies that pursue efficiency through AI are optimising themselves for precisely that space: the combinational middle. When everyone can use the same tools to produce the same level of quality, price becomes the deciding factor. Agencies cannot win a lasting price war in a commodity market. At the same time, they lose something that is difficult to recover.

Research shows that skills atrophy when they are no longer exercised. Dorothea Winter calls this de-skilling. A study published in The Lancet Gastroenterology & Hepatology measured the effect among doctors. Endoscopists who had spent several months using AI-assisted polyp detection achieved a lower detection rate when subsequently working without AI than they had before the technology was introduced, falling from 28.4 to 22.4 percent.

A muscle that is no longer used grows weaker. Anyone who continually outsources creative judgement eventually loses the ability to exercise it. Medicine can already demonstrate this effect empirically. Agencies are beginning to experience it for themselves.

Dorothea Winter identifies three capabilities that matter in the age of AI: judgement, selection and authorship.

Judgement is the least conspicuous of the three. It means preserving your own sense of what is good and what is not, regardless of how much AI contributed to the result. A strong brief does not need someone to manage AI outputs. It needs someone who knows what they want and why.

Then comes selection. AI generates options. Choosing the right one from twenty concepts requires a judgement the AI cannot provide, and one that only someone who understands the craft can make.

Finally, there is the most uncomfortable question: authorship. Who stands behind the work, and truly means it? Not who used the tool, but who takes responsibility for the outcome and is prepared to stand by it. That is the question agencies cannot delegate to AI.

Output, production and first drafts can be delegated. Conviction, responsibility and judgement cannot.

Thomas Strerath has expressed the same idea in different terms: an agency’s intellectual signature can hardly be separated from the people who carry it. What makes an agency distinctive lies in the people who exercise judgement, and actively want to do so, not in the tools they use.

And that is precisely the problem with the middle. Its work is not bad. It simply lacks someone willing to stand behind it with conviction. The question is what kind of organisation makes that possible in the first place.

In corporate structures, conviction eventually becomes a vote and judgement becomes a process. What is to be created is assessed in advance for risk, consensus and compatibility with other client relationships. This does not happen because of bad intentions. It is the logic of the system. Agencies that operate this way cannot develop an intellectual signature. They can only manage what is already acceptable to the majority.

Clients, meanwhile, want two things at once: the agility of owner-led, independent teams and the reassurance of scale. That is a genuine dilemma. Size alone will not solve it. What matters is whether an agency has the structures in which authorship can emerge at all.

The bifurcation Lucy Jameson describes is not a prophecy that will simply befall agencies. It is a decision every agency makes through whom it hires, what it turns down and which clients it chooses to work with.

Those who refuse to choose have already chosen.

And for those who choose the middle, Lucy Jameson has already told us what happens next.