Omnicom cut 4,000 jobs in H1 2026. Nearly a fifth of their combined headcount.

10,000 positions eliminated total since the IPG merger. Three creative network brands retired outright. Headcount dropping from 128,000 to roughly 105,000.

The financial press described it as "eliminating overlap." That's accurate. Overlap IS the holdco model.

Account teams at the holding company, duplicated at the network, duplicated again at the specialty shop underneath. Stack three levels of the same delivery function across 128,000 employees and the math eventually produces a spreadsheet nobody wants to write. The merger forced the spreadsheet.

What's worth noting beyond the headline numbers is what the reorganized entity looks like after the cuts settle. Our read: a structurally similar operation with better unit economics on paper. Same delivery architecture, fewer seats. The brand-retirement announcements - three creative networks retired outright - are the clearest signal the holding company isn't rebuilding into something architecturally different. It's compressing into a tighter version of what already existed.

That's a rational choice for a public company managing headcount costs. It doesn't address the underlying design question.

The design question: at what headcount does a performance marketing operation produce its best work?

Once AI systems carry the analyst work, the build work, and the reporting work, the headcount math changes earlier than most holdco planning assumptions ever built for. The analyst seat that read last quarter's data and wrote this quarter's brief is now a pipeline. The reporting seat that exported spreadsheets every Friday is a query. Each one was a headcount line in the holdco model. None of them is anymore.

Every duplicate function Omnicom eliminated - the parallel account team, the redundant creative layer, the copied planning function - is a function that, in a system-first operation, never needed to exist in the first place.

The analytical workload that a traditional setup staffs across a 10-person account team runs as one agentic system in ours. We didn't build that to replace people we'd been planning to hire. We built it because the alternative was the holdco model: adding headcount to solve delivery problems that infrastructure could solve instead.

The Omnicom-IPG restructuring is a forced compression of a pre-AI design. Every function being cut was built for a labor market where headcount was the only available answer to volume. That labor market no longer exists.

The agencies that survive the next compression cycle are the ones that designed for AI-era operations from the start - or the ones building that design now, before their own spreadsheet gets written.

The cuts make the quarter's numbers work. They don't change the architecture underneath, which is the part that determined the headcount in the first place. When the infrastructure answers the volume question, the org chart stops being the lever you reach for.

Every agency website runs the same 4 words. "We drive real results."

It used to take a track record to write a sentence like that. Now any team can generate a polished case study PDF, a testimonial wall, and a slick results page before lunch. The words got cheap. The trust they were supposed to carry left with them.

So the claim stopped doing any work in the buyer's head. What a prospect still can't fake is a live dashboard with their own login - the screen where the campaign names and the spend are real and the line either climbs or it doesn't.

We post ours. 3.1x+ blended ROAS across 20+ brands. If it were inflated, anyone with the login would catch it in a minute.

The "I shut my agency down" post is everywhere this quarter. Most of them bury the real reason.

The execution-only layer is getting commoditized. That's the actual story. Clients are pulling work in-house using AI tools that do in 4 hours what a retainer line item used to justify across 3 weeks. The landing page development add-on, the copywriting package, the weekly reporting deck - those are the things disappearing.

The client relationships don't go with them. The margin on execution does.

What survives is the thinking clients can't replicate with a $30/month tool subscription. Strategic diagnosis. Pattern recognition across a portfolio of accounts. The system architecture that makes AI output actually perform at the account level instead of sounding like everyone else's AI output.

We've watched this play out across the accounts we run. The clients who've been with us longest aren't paying for execution volume. They're paying for the calibration layer that makes the execution win. The brief system that reaches a winning creative angle in a fraction of a standard research cycle because it starts from that account's own conversion history. The architecture that keeps naming conventions, conversion actions, and bid logic consistent across a portfolio so no single account drifts when one campaign changes.

That calibration layer was always the actual product. Execution was just the visible deliverable it hid behind.

The agencies posting "I shut this down" built identities around delivery speed and volume. Both of those are now cheaper with AI than with a team of junior operators. What remains is judgment - and judgment requires the kind of institutional memory that takes years to build across real accounts with real spend.

The accounts we've run for 18+ months are the ones where the calibration layer compounds. Month one, the AI systems run the standard build. Month six, they're running a build informed by 6 months of that client's winning and losing patterns. Month twelve, the creative brief starts from a different place than any new agency could start from, because the system has read every test that account has ever run.

That's what clients stay for. The AI can generate the ad. The institutional memory determines which ad to generate.

So the "I shut my agency down" essays are reporting something real. The execution layer those businesses were built on did collapse. The agencies still standing are the ones where execution was never the actual product, just the part of it the client could see.

Ramp posted a "Vibe Growth Marketing Manager" role this week. The job title is already the review.

Vibe marketing is the idea that you describe what you want and AI ships it. Campaign, copy, creative, landing page, all from a prompt and a feeling. For the first two weeks it genuinely works. The output looks professional. The dashboard fills up.

Then day 90 arrives. The angle that felt clever has fatigued, the audience the prompt guessed at turns out wrong, and nobody wrote down why any of it was built that way. There's nothing underneath to diagnose from.

Roughly 90% of marketers use AI daily now, so having it in the workflow is table stakes. The system that decides what to feed it is the part that holds up at 90 days.

OpenAI showed up at Cannes Lions this week. Their CRO said they're "clearly in the advertising business now."

Same week: Meta launched Brand Memory on June 23 - an end-to-end AI creative tool that generates, targets, and optimizes ads inside Meta's own ecosystem. Award entries at Cannes are down roughly 25% after last year's AI-faked-results scandal caught up with the industry's credibility.

Three things in the same week, describing the same shift: the AI labs are moving into the media business, the platforms are absorbing the creative stack, and the showcase work that used to validate agencies is now under suspicion.

The question this raises for anyone running a performance agency is more uncomfortable than the platform announcements suggest.

When Meta owns the AI that writes your ad, places your ad, optimizes your ad, and holds the data that evaluates your ad - the agency's role in that stack narrows. The short answer in 2024 was "strategy and creative judgment." That answer is getting shorter.

What holds across every platform consolidation cycle is accountability.

The agency signs the contract. When results drop, the agency answers for it. Platform AI carries no contract liability. That's the specific thing clients are still paying for when the platform AI handles more of the build: someone who owns the outcome.

Platform AI is genuinely good at production and optimization within its own ecosystem. What it hasn't replaced is the operator who reads 6 months of a client's customer service tickets, runs 40 presell page variants to find the angle that actually converts, and knows why the previous version failed. That institutional memory is what determines whether the AI-generated creative is the right AI-generated creative.

The brief system does the research build. The operator carries the account history. The platform AI is fully capable of the first job. It can't do the second one yet, because it hasn't sat in the account long enough to know which past test the current decision should be weighed against.

The Cannes award-entry drop (roughly 25%) is what happens when an industry uses AI to generate showcase work without building the accountability layer around it. The platform shift - OpenAI entering ads, Meta closing the loop on creative production - is the infrastructure following the same pattern at scale.

The agencies that hold the accountability the platform AI can't carry are the ones that survive this consolidation. Volume production is a platform capability now. Accountability for the account's actual outcome is what the external partner still owns.

OpenAI standing on the Cannes stage is the signal that this stopped being a forecast. The generation layer and the distribution layer are consolidating into the platforms. The work that doesn't consolidate with them is the work where someone signs their name to the result.

Someone spent millions of Claude Code tokens building their 10th app this year. No customers.

That post got passed around because everyone recognized the trap. AI collapsed the cost of building to almost nothing, so everyone builds. The product stopped being the scarce thing. Attention is.

Distribution was always the hard half. It just used to hide behind the months it took to ship anything, which made building feel like the real work. Now building is a weekend, and the channel is the whole game - which stings, because a channel is the one asset AI can't generate for you. You earn it slowly or you rent it expensively.

The operators who saw this two years ago have an audience now. Everyone else has another MVP.

The "90-person agency burnout" posts are everywhere. We tracked the math on going the other way.

The dark side of a large agency reads like a logistics problem: payroll weight that doesn't flex when a client churns, 8-hour call days across account managers and clients and internal reviews, hiring cycles that take 3 months to produce someone who needs 3 more months to contribute, churn that resets the revenue clock every quarter. We've watched this list grow in friends' businesses. It's long.

Ours is 2 lines.

Ruslan runs Google Ads operations and the AI systems architecture. Andrey runs Meta. Alex runs retention via Klaviyo. Three operators, $10M+ a month in client ad spend, 20+ brands, 3.1x+ blended ROAS. Zero account managers between us and the accounts. Zero layers between the client's budget and the people making the actual decisions on it.

We chose this deliberately. It didn't look like the obvious choice in 2022.

The traditional agency growth path goes like this: hit capacity, hire account managers, hire junior buyers, build a management layer to supervise the junior buyers, sell the additional capacity as proof of scale. The margins compress as the headcount grows. The founder moves from accounts to management. The work that won the original clients is now the work the junior team does, supervised from a remove.

We built systems instead.

The AI Creative Brief System that Andrey uses to produce 250 Meta ads a week from one seat. The Google Ads AI Agentic System that holds naming conventions, conversion actions, and bid architecture consistent across every account at once, instead of a manager checking each one by hand. The reporting pipeline that replaced the 4-hour-per-week spreadsheet export.

The systems are why the burnout list is short. They're also why the DFY retainer (tegra.co/work-with-us) is founder-run in a way that's actually true. When we say founder-run, it means Ruslan is on Google, Andrey is on Meta, Alex is on retention. The accountability structure is the three of us, not an account manager who briefs a junior buyer who runs the actual account.

We're 3 people by design. The constraint we kept is the one that keeps the senior operator on the account. That's what the systems made possible - removing headcount without removing expertise from the work.

The dark-side list from the 90-person agency posts is real. We see it in the clients who come to us from those setups. What they describe is consistent: they want a senior operator on the account, owning the decisions. The systems are how we keep that true at the volume we run without the organizational weight that usually rides along with it.

Nearly 70% of Google searches ended without a click in early 2026. Over 90% inside AI Mode.

Read that against what most SEO retainers still sell: we'll rank you for the how-to keywords. That deliverable assumes the user clicks through to your page. In a zero-click result, the answer gets assembled on the results page and your brand shows up as a citation, not a visit.

So the work changed underneath the contract. The question moved from "are we ranked" to "are we the brand the model quotes when it writes the answer." Different content, different structure, different proof signals to earn.

Brands cited in AI answers see +35% CTR on the clicks that still happen (Seer Interactive, 2026). Keyword ranking still matters for the searches that click. That's a shrinking share of them.

Every agency is pitching a $25K/mo AI offer now. Almost none show what happens after the pitch deck.

The offer structure is straightforward: AI handles production, the agency provides strategic direction, the client gets volume and speed at a fraction of a traditional team's cost. Hard to argue with as a pitch.

The problem arrives in month two. The founder is still in every creative decision, every account change, every client call. A delivery model that requires the founder to personally approve each output is structured like a high-end freelance arrangement, regardless of how the pitch frames the AI angle.

The delivery layer is the boring part nobody shows in the case study.

Here's what ours looks like.

The AI Creative Brief System generates a research-backed brief from customer intent signals. It's a structured document that tells the creative build exactly which angle, which audience framing, which proof point, and which format the data says is most likely to win. Andrey runs that system across every Meta account we manage. One operator. 250 ads a week.

The Google Ads AI Agentic System handles the multi-account architecture. Naming conventions, conversion action consistency, audience taxonomy, bid architecture - held consistent across every account change by the system, instead of a manager re-checking each account by hand. Ruslan runs that layer across the whole portfolio.

Alex runs retention across the same client base via Klaviyo. 15+ email campaigns monthly as part of the DFY retainer output.

The output (ads, presell pages, email campaigns) is where most AI-agency pitches stop. We built the input layer - the brief system, the architecture standards, the audit protocol - because that's what determines whether the output runs consistently without the founder reviewing every piece.

One system, three operators reviewing the output instead of building it. 250-1,000 ads, 10-25 presell pages, and 15+ email campaigns monthly. Live in 3 days. That's the delivery layer behind the pitch.

The $4,997 Google Ads AI Agentic System (tegra.co/store/google-ads-ai-agentic-system) is the self-deploy version of that infrastructure - for the boutique agency owner or PPC freelancer who wants to run the same architecture without building it from scratch. The DFY retainer (tegra.co/work-with-us) is the version where we run it for you.

The pitch is easy. Showing the operational layer is rarer. We show it because the operational layer is the actual product.

Smart agencies are building creator networks, not in-house content teams.

Variable cost, real distribution, no payroll carried between campaigns. The pegs sound absurd until you run the math. A Pilates studio reportedly paid one of its own members $18K a month just for footage. Operators are running distribution networks of 5,000+ creators.

Both look insane next to a salaried two-person content team, until you notice the studio pays only when content performs and the salaried team gets paid either way.

That's the shape of a lean agency in 2026. The content function moved from a fixed cost you carry to a variable cost that tracks output. The algorithm already favors the native creator clip over the polished in-house spot, so the cheaper model tends to be the better-performing one too.

7 domains we audit on every account before we touch a single bid.

We run client ad spend across 20+ DTC brands, and the audit is the same seven domains every time.

Written down, so any operator on the team runs it identically.

Here's what each domain checks:

• Conversion tracking: is the pixel firing on purchase, deduplicated, and reconciling to the back end
• GMC setup: feed health, item-level disapprovals, and the attributes that quietly cap impression share
• Targeting: audience signals, intent tiers, and where spend physically lands versus where the brief assumed
• Bidding strategy: the right strategy for the data volume, since tROAS switched on too early is the top reason scale stalls
• Landing pages: query match, load speed, and whether the mobile experience holds up where most clicks are
• Account setup: conversion actions, naming conventions, and exclusions consistent so downstream data is trustworthy
• Campaign structure: brand and non-brand separated, intent unblended, and whether three hero SKUs eat the budget while hundreds of products sit at zero impressions

This is codified for the same reason we codify everything: write the process once, then let the team and the tooling run it the same way on every account.

Three of us reviewing identical output beats three of us improvising.

The pattern across almost every account we audit: it breaks on the first two domains.

Tracking and feed.

If an agency runs your account, ask to see their audit protocol - not the findings, the written checklist. "Every account is different" means they improvise on your money.

Optimizing before auditing just spends faster in the wrong direction.

We lost a client by giving correct advice. Their landing page was the problem.

We diagnosed it. We didn't fix it.

Three months after the audit, the page was still broken.

Cold traffic still converting at 1.5%.

Client churned.

That failure changed how we build retainers.

We ship 10-25 presell pages per month inside our DFY retainer now.

New accounts go live in 3 days, because we built the delivery system before making the promise.

Cold traffic from Google or Meta arrives with no brand relationship and no reason to trust.

A product page is built for buyers who already want the product.

A presell page does the belief-building work in between

- handles objections
- establishes proof
- gets the visitor warm enough that the product page can close

Without that layer, you're paying for clicks your funnel isn't built to convert.

One client scaled past their own inventory capacity after we rebuilt the funnel architecture.

Their exact words: "scaling so much we're scrambling for stock again."

The boutique agency failure mode isn't bad diagnosis.

It's good diagnosis with no delivery infrastructure to act on it.

Diagnosis gets you nodding clients.

Delivery is what keeps them - we learned that one churn too late.

Three of us run $10M+ a month in ad spend across 20+ brands.

The staffing math says that's a 15-person shop. The gap isn't heroics - it's 147 internal systems that hold what agencies usually store in employees.

The "company brain" idea is having a moment: a business's real memory lives scattered across calls, Slack, dashboards, and people's heads, and whoever consolidates it wins.

For agencies, the future tense is wrong. This has been our operating math for two years, and we can put a ledger on it.

What the brain replaces, concretely:

The account manager's memory of why a campaign is structured that way - written into the audit system that checks the structure nightly.

The senior buyer's instinct for a dying feed - encoded as checks that run before anyone's coffee.

The founder's pattern-recognition across clients - a changelog every system reads, so brand #19 inherits what brand #4 taught us.

The onboarding doc nobody updates - dead, because the system IS the doc, and it can't drift from practice when it runs the practice.

One of them, anatomized - the overnight feed check:

• Trigger: every account, before dawn
• Input: the product feed plus yesterday's diff
• Output: out-of-stock flags and price mismatches, drafted as fixes
• Approver: whoever owns the account, before their first call

Here's the part agency owners resist: every one of those systems started as a sentence someone was tired of repeating. That's the whole method.

The brain isn't built in a quarter. It accretes one written-down decision at a time, and the compounding is brutal for anyone who starts late.

What it doesn't fix: judgment calls, client trust, and taste. We still argue about those daily - the brain just stops us paying salaries to remember things.

The write-it-down test, for any agency owner: next time a client question gets answered from someone's memory, ask what happens when that person is on a plane.

If the answer is "we wait," the agency's brain is on that plane too - and the version of you that writes things down is two years from replacing you.