The agency exodus isn’t a talent problem. It’s the model cracking.

If you run a UK agency, the Guardian piece should make you pause, mainly because a big chunk of the departures weren’t layoffs; they were resignations. They chose to go, and that changes how you read the rest of the numbers.

The IPA’s latest census puts total employment across member agencies at 24,963 as of 1 September 2025, down from 26,787 the previous year (a 6.8% decline). The biggest hit was in creative and other non-media agencies, where employment fell 14.3% (from 14,775 to 12,659). The under-25 group dropped 19.2% (from 3,632 to 2,936).

And the Guardian reports that the “exodus” was driven by younger workers, with over half of departures voluntary. It also notes that agencies expect more cuts that’ll get blamed on AI in 2026, and that advertised jobs fell sharply. 

You could put it down to a rough year. But when younger people leave in those numbers, and when resignations are a meaningful chunk of the exits, it usually means the deal doesn’t feel worth it anymore. Not just pay. But the whole deal: the time, the pace, the friction, the feeling that you spend more energy navigating the system than doing work you’re actually proud of.

That’s what’s going on here.

People will say this is AI taking jobs. I don’t think that’s quite it. I think AI is speeding up the work, and once that happens, the old setup just feels slow.

Not because AI is perfect. It most definitely isn’t. AI isn’t a magic button. You still have to know what you’re doing.

But it’s fast.

It can give you ten options before you’ve finished your first coffee. It can iterate, rewrite, restructure, summarise, create variations, and get you to a workable first pass much quicker than traditional workflows.

Once the work speeds up, people stop tolerating all the layers around it.

Because if you look closely, many of the network models exist for internal reasons rather than client outcomes.

Layers exist because layers have always existed. Meetings exist because you need meetings to keep handovers moving. Roles exist because big organisations need a “middle” to coordinate the machine. None of that is bad; it’s just what happens when you scale headcount and try to keep the wheels on.

The problem is that clients don’t pay for internal complexity; they pay for outcomes.

They’ve put up with it because they assumed that’s what it takes to get good work. When output gets quicker, people have less patience for the nonsense. You start hearing different questions on client calls, especially when there’s pressure:

• “Why are there eight people here?”
• “Why does everything take two weeks to come back?”
• “Why can’t someone tell me what’s happening this week?”
• “Why am I paying for coordination rather than progress?”

That’s when “big” stops feeling reassuring and starts feeling bloated.

The Guardian notes that 24% of agencies expect to cut jobs directly due to AI in 2026, and that job postings dropped significantly.

Some cost-cutting will be unavoidable. Some roles will change, and that’s just a reality. But if your entire response is “same model, fewer people,” you’ll keep the wrong parts and lose the right ones. Partly, because you’re not removing friction, you’re removing capacity.

You’re basically asking fewer people to run the same machine. And when you do that, it doesn’t get “leaner”, it just gets more tense. The pressure goes up, the good people burn out quicker, everything starts leaning on a couple of heroes to keep it moving, and delivery gets shakier because there’s no slack left when something changes (and something always changes).

It also sends a message internally that talented people pick up instantly: we’re not redesigning the work, we’re just shrinking the human part of the machine.

The best people don’t leave because they’re scared of AI.

They leave because they’re fed up with the pointless stuff. Chasing updates, sitting in calls that only exist because nobody properly owns the outcome. Spending hours building decks about decks and fixing the same avoidable problems every single week because the system never changes. Good agency people will graft; they can handle pressure, deadlines, client chaos, all of it. What they can’t stand is wasted energy. And that’s what the “exodus” story is really pointing to.

Most agencies don’t have a talent problem; they’ve got a shape problem. As they scale, a whole layer of in-between work builds up just to keep the place moving: people writing updates about updates, PM turning into status-chasing, briefs getting watered down every time they’re handed over, and meetings happening simply because ownership isn’t clear. Clients don’t want to buy that. They want outcomes.

Your best operators don’t want to spend their careers administering it. And when AI speeds up production, the “in-between layer” becomes more visible, not as a moral issue, but as an economic one. Clients start asking why they’re funding internal complexity.

That’s what’s changed. Not “AI writes copy now.” It’s “clients stop paying for how complicated your organisation is.”

When people say “the future is independent,” some hear chaos. That isn’t what I mean. I don’t think the answer is one mega-network swallowing everything.

And I also don’t think it’s a bunch of random contractors glued together at the last minute, either.

The model that makes the most sense is independent agencies working collectively, with a shared approach to delivery, shared standards, shared tooling, and modular teams that form around the problem and stand down when the job is done.

Independence keeps people sharp. Collective delivery gives you scale. And the collective model has one huge advantage over the traditional network structure:

It lets you build capability without carrying permanent bloat.

You pull in the specialist when they’re needed. You don’t keep them on payroll just to justify a department’s existence. You don’t add layers to coordinate layers. You build the team around the work, not around the org chart.

Clients like it because it’s just clearer. Fewer handovers, fewer “we’ll come back to you” loops, quicker decisions, and ownership stays with a person instead of disappearing into a department. And the team usually prefers it too, because it feels closer to the point, less politics, less theatre, more craft, more responsibility, and a clearer link between the effort you put in and what actually moves.

You can see even the biggest groups responding to the same pressure in their own way. Reuters reported that WPP plans to bring major creative agencies (including Ogilvy, VML, and AKQA) under a “WPP Creative” holding structure to simplify the client offering while preserving agency brands.

Whether you love that move or not, the signal is obvious: complexity is becoming the problem.

People talk about it like it’s a choice between going all-in on AI and devaluing humans, or ignoring it and sticking to “craft”.

But real agencies don’t live in extremes. They live in delivery.

The practical version is: stop arguing about it and decide where it helps. Let AI handle the grind, first passes, versioning, admin, basic synthesis, the bits that quietly eat the day.

Keep humans on the parts that actually need a person, the judgment calls, the trade-offs you can’t spreadsheet, the accountability when it’s gone a bit sideways, the trust-building with clients, and the awkward but necessary conversations when something isn’t working. Basically, the decisions that have real consequences, and the kind of strategy that still holds up once it hits real life.

If you’re an agency leader, that’s the point: AI isn’t your strategy, it’s a tool. The strategy is to redesign delivery around what clients actually value and what talented people actually want to spend their days doing.

That under-25 drop should worry the industry for a simple reason: if the juniors disappear, you’re not just short on bodies now, you’re short on seniors later. Nobody wakes up one day with great judgment. You build it by doing the work, making mistakes, being coached, and getting reps. 

So if early-career hiring shrinks and younger people decide agency life isn’t worth it, you’re quietly removing the next wave of delivery leaders before they’ve even had a chance to grow.

And you won’t feel it straight away. It usually hits three to five years down the line: delivery gets patchy, clients can tell something’s off, the bench is thin, and the same handful of people end up firefighting everything.

The answer isn’t “keep juniors because it’s nice.” It’s to build a better entry path for a different kind of agency world, one where speed is higher, tools are different, and the valuable skills are changing.

Collectives can actually help here if they take training seriously, because training doesn’t need to live inside a single giant organisation. You can run shared apprenticeships, shared standards, and shared playbooks while still keeping delivery modular and lean.

‍

No big strategy deck. Just changes you can actually make.

1) Remove one layer between the client and the work

Pick one client and remove one handover step for a month. If you want to push it, remove the ‘translator’ layer and let the people doing the work speak directly to the decision-maker, with a senior person there to keep it tight. What usually happens is that things move faster, the brief doesn’t get mangled along the way, and the team stops second-guessing.

2) Make accountability a name, not a department

If the internal answer to “who owns this?” is “Account Management,” you’ve got a client risk. Clients want a person who can say: “Here’s what’s happening, here’s what we’re doing, here’s what we’re not doing, and here’s why.”

3) Put AI in the workflow, not in a training deck

Most agencies treat AI as “a tool people can choose to use.” That creates inconsistency fast. Decide where AI belongs in your standard delivery process. What gets drafted, what gets versioned, what gets summarised, what gets checked, and what stays human-only. Then make that the default, so you’re not relying on individual enthusiasm.

None of this requires a big transformation programme; it just requires decisions.

‍

If AI made delivery 30% faster tomorrow, would your business get stronger… or would it wobble? If it wobbles, that’s normally a sign your model relies on hours staying high for it to make sense. So the problem isn’t the tool, it’s the setup.

Second: if a client asked, “Who’s accountable for this?” would you name a person… or a department? If your answer is a department, clients feel the gap immediately.

Collectives win because accountability stays human even as delivery speeds up.

The network agency model isn’t dated because big agencies are bad. It’s dated because the shape doesn’t match how value gets created now.

The agencies that do well from here won’t be the biggest. They’ll be the ones that stay lean and specialist most of the time, pull in extra capability when it’s needed, keep ownership clear (a name, not a department), and invest in training so juniors actually become the next wave of seniors as the work changes.

They’ll use AI to make delivery cleaner and faster, not just cheaper. That’s how you stay efficient without losing what makes the work good, and how you scale without layering on unnecessary weight. In other words: technology should improve the way you deliver, not turn the whole business into a headcount exercise.

If you run an agency, I’m genuinely curious:

What part of your org chart exists mainly because work used to be slower?

‍