35,000 Agency Jobs Cut — Here's the Team Structure That Actually Survives

Short answer: The holding companies just showed you the answer. WPP cut 12,000 positions. Omnicom is targeting 15,000. Copywriting job postings in agencies went from 1-in-6 to 1-in-30. The junior execution layer is being hollowed out — while demand for senior strategists is climbing. The agency team that works now has three seats: the Operator (runs the machine), the Closer (fills the contracts), and the Taste Layer (makes the machine get better every week). The rest is the machine's job.

The news hit my feed the same week I watched an agency owner on a call try to explain to his team of eight why three of them spent the entire week doing work a system could handle in an afternoon.

He knew it. They probably knew it. Nobody was saying it out loud.

Then WPP made it loud. Twelve thousand positions cut since early 2025. A new CEO from Microsoft running a restructuring programme targeting half a billion pounds in savings. Her quote landed like a verdict: "AI automates the ordinary and massively elevates the extraordinary."

She's not wrong. She's describing what already happened at the scale most agency owners actually operate at.

What's actually happening to agency teams right now?

The holding company cuts are the headline. The structural shift underneath is the story.

Omnicom is targeting 15,000 total position reductions after absorbing IPG. Dentsu cut 3,400 internationally. Across the major holding companies, more than 35,000 positions have been eliminated or targeted in the past 18 months.

But the number that matters isn't the total — it's the shape of the cuts.

Copywriting job postings in agencies went from one in six to one in thirty, according to The Drum. Gartner's CMO Spend Survey found 23% of agencies already reduced junior copywriting roles in 2025, with another 31% planning more cuts. An analysis of 8,000 content marketing job listings found mid-level postings dropped more than 70% — while senior-level postings increased more than 300%.

The junior execution layer is being hollowed out. The senior strategic layer is in higher demand than ever.

D&AD's 2026 creativity report, based on conversations with 197 creative leaders across 30 countries, named the pattern directly: "The entry-level jobs that trained future leaders are disappearing. AI now does the repetitive work juniors used to learn from."

Forrester doubled their agency job loss prediction — from 7.5% over eight years to 15% in 2026 alone.

This isn't a recession. It's a restructure. And if you run a small agency, the question isn't whether it hits you — it's whether you design the restructure yourself or get restructured by a market that won't wait.

What are the three seats that actually matter?

When I stripped our agency down and rebuilt it, three roles emerged. Not "roles AI replaces" — roles the machine makes possible.

Seat 1: The Operator

This is the person who runs the machine. Not the ads, not the funnels, not the emails. The machine.

In the old model, you had a media buyer manually building campaigns, pulling reports, adjusting bids, writing ad copy, requesting creatives, then summarizing it all in a Monday meeting nobody wanted to attend.

Now the machine produces 13 ad concepts and 6 finished creatives in one afternoon. It monitors every account before anyone wakes up and pre-queues the decisions. It publishes content autonomously — we've shipped more than 70 posts without a content calendar meeting ever happening.

The Operator doesn't do those things. The Operator makes sure the machine does them correctly. They watch outputs. They catch drift. They feed new data in. They notice when something breaks before the client does — like the time we had a silent delivery failure for 9 days because a single missing character meant everything looked fine while buyers got nothing.

That failure taught me more about this role than anything else: the Operator's job is to never let the machine lie to you.

Seat 2: The Closer

This is the one thing AI genuinely cannot do: sit across from a human being and close the deal.

Not "nurture." Not "send a sequence." Close.

Discovery. Objection handling. Reading the room. Knowing when to shut up. Knowing when the prospect needs you to say "here's what we'll do in the first 14 days" instead of pitching another feature.

Every agency needs a Closer. What changes is that the Closer now walks into every call armed with intelligence the machine already gathered — account analysis, competitive landscape, performance benchmarks — instead of spending two hours prepping a deck the night before.

We've generated over 2,700 booked calls at a best cost per booked call of $27.22. But none of that matters if the person on the other end can't close. The machine fills the calendar. A human fills the contract.

Seat 3: The Taste Layer

This is the seat nobody talks about — and the one WPP's CEO was pointing at when she said AI "massively elevates the extraordinary."

The machine produces output. A lot of it. Fast. But output without taste is noise at scale. That's exactly what the D&AD report warned about — AI-declared award entries actually won at lower rates than human-led work when the human judgment layer was thin.

The Taste Layer is the person (usually the founder) who makes the corrections that make the machine get better every week. Every rejected headline, every rewritten hook, every "this doesn't sound like us" feeds back into the system so the machine never makes that mistake again. It compounds. Week over week.

I learned this the hard way. We built a $1,497 product page in a single session — idea to live checkout. Fast. Impressive. And the copy was weak, because it skipped the review process. The version that went through the full pipeline — scored by independent reviewers, judged blind, assembled into a hybrid — was unrecognizable. Same product. Same afternoon. Completely different conversion potential.

The Taste Layer is why an AI-native agency isn't just "an agency that uses AI tools." Tools don't have taste. The machine you own does — because your taste is baked into it.

What does this look like in practice?

The industry data above isn't theory to me — it's what I already lived through on the way to this answer.

Right now: it's me and the machine, running client businesses across multiple niches.

• Me: Taste Layer + strategy + closing. I make the corrections. I'm on the calls. I set the direction. The machine handles the rest.
• The machine: Creative production. Content publishing. Account monitoring. Performance reporting. Quality checks. Campaign builds. Morning briefings. Everything that used to require a team of specialists doing manual work all day.

We went from a model where scaling meant "hire another media buyer" to one where scaling means "feed the machine more data." The machine that monitors one client's account before I'm at my desk monitors ten the same way. The system that runs Meta ads on AI for one account does it for all of them.

Millions in ad spend managed. Client revenue from $30K to $105K in six months. Not with a team of 20. With a structure built around what humans actually need to do — and what they don't.

Why is waiting to restructure the most expensive decision you can make?

This is the TIME argument, and the holding companies just made it for me.

Every month you keep your current org chart — the one designed for manual execution — AI-native competitors compound past you.

The machine gets smarter every week. Every correction, every client interaction, every campaign result feeds back in. Week 1 the machine is clumsy. By week 12 it knows your accounts well enough to kill a $3.02 ad after 40 impressions and zero clicks — same day, no meeting, no lag.

WPP's restructure will take years. Omnicom's cuts are projected through the end of 2026 and beyond. You don't have to move at holding company speed. You can restructure in weeks — if you commit to the rebuild instead of dabbling.

The restructure isn't a layoff. It's a rebuild. And right now it's the most important thing on your calendar.

If you want to see exactly how we built this — the system, the structure, the economics — I put everything into a $27 playbook. No theory. Just the machine we actually run, broken down step by step.

Frequently Asked Questions

How many people does an AI-native agency actually need?

The pattern we see — and run ourselves — is 3-4 core people plus the machine. The D&AD report and the Semrush job-listing data both point to the same conclusion: headcount stops scaling linearly with client count. One Operator can manage the machine across multiple clients because the system handles execution; the human handles exceptions and judgment.

Does restructuring for AI mean I should fire my team?

Not necessarily — but their roles change completely. The data is clear: the junior execution layer is being automated, while strategic roles are in higher demand. A media buyer who manually builds campaigns is doing work the machine handles. A media buyer who operates the machine and makes strategic decisions is more valuable than ever. The seat changes shape; the person in it might stay if they're willing to evolve.

What happens to quality when a machine handles most of the output?

It depends entirely on whether you have a Taste Layer. Without one, you get quantity — lots of mediocre output, fast. D&AD's report flagged exactly this: AI-declared award entries won at lower rates than human-led work when the judgment layer was thin. With a Taste Layer, every correction compounds. We tested this directly: a single-pass page versus one that went through multiple rounds of review. Same product, same day — night-and-day difference.

How long does it take to rebuild an agency team around this model?

In our experience, the core takes 60-90 days to get running, then it compounds from there. The dangerous move is trying to do it gradually while maintaining your old structure in parallel — you end up paying for both and getting the benefits of neither. WPP is spending years and half a billion pounds on their restructure. A small agency can do it in weeks because there's less to dismantle.

What's the difference between this and just using AI tools?

Tools don't compound. A ChatGPT subscription gives you the same output today as it did last month — no memory of your clients, no accumulated judgment, no permanent rules from lessons learned. The machine you own reads every account, applies every correction you've ever made, and gets harder to replace every month it runs. That's why your clients can't rent the same thing you're running — they'd need the taste, the data, and the compounding.