UNDERTOW 007: Guessing at the Speed of Confidence
A provocation for anyone running a creative organization on someone else’s clock
UNDERTOW is an infinite report. A limitless USB stick for cultural intelligence. A living container that doesn’t finish, only accumulates, and keeps... growing. The concepts travel when readers use them in rooms I’ll never enter. The only requirement is that each piece is honest to what I’m seeing right now.
Everyone in the industry is arriving at the same answer at the same time, and nobody seems nervous about that.
The answer is taste. You’ve heard the argument by now, and it’s elegant the first time you hear it: AI commoditizes production, production was the revenue model, so what survives is the thing that can’t be automated. The human ability to know what’s good and refuse what isn’t.
This is true. It is also the most comfortable possible version of the truth.
Taste is what every displaced creative leader says is the answer, because taste is the thing they have. If the answer is taste, then the senior creative just needs to be recognized for what they already are. No structural change required. No new model necessary. Just: see me, value me, pay me.
(I know because I’ve been that person. I’ve made that argument. It felt true every time I said it, which is exactly why I stopped trusting it.)
Here’s what I started trusting instead.
Every creative organization runs two kinds of processes at once, and only one of them can defend itself. The first kind lands fast: an A/B test that picks a winner by Friday, a media plan judged by its first weekly report, a feature that shipped because it tested well in a sprint demo and was quietly removed six months later, a campaign that launched to client applause and was forgotten by the audience before the invoice cleared.
The second kind lands slow: the brand platform that takes eighteen months before anyone can tell if it reshaped perception. The campaign idea that tested poorly in research and became the most awarded, most effective work in the category a year later.
Every optimization model in the history of business has the same bias. It favors the fast kind. It favors the feedback loop that lands in days, not years. And for forty years, this industry has been installing fast-feedback metrics as the operating system of a slow-feedback craft. The slow process couldn’t defend itself in the language the metrics spoke. So the fast metrics ate it alive.
The creative director who killed the good campaign at midnight wasn’t exercising taste. Taste implies a snap judgment — good or not good. What she was exercising was something slower and harder to name.
She had lived inside the problem long enough to know the solution wasn’t finished. Not because it was bad. Because she had developed a relationship with the problem that was deeper than the relationship the work had with the problem, and she could feel the gap.
That kind of knowledge doesn’t come from talent. It comes from time. From the willingness to stay inside a problem past the point where a competent answer has already appeared, because you have enough experience to know that competent and finished are not the same thing. Call it... duration.
The scarce resource is not taste. Taste is real, and rare at the highest level, and everyone who has it deserves to be valued for it. But taste without duration is guessing, done at the speed of confidence. You’ve seen it: the brilliant creative leader who arrives, makes bold calls in the first ninety days, and is gone in eighteen months because the bold calls didn’t hold up. Not because his taste was wrong. Because he hadn’t been inside the problem long enough for his taste to become judgment. The scarce resource is not the taste. It’s the organizational willingness to fund the time it takes for taste to become operational, and to protect that time from every model that will, correctly, identify it as waste.
The most valuable thing in the building is the thing that looks most like waste.
That is the structural argument. Here is the evidence.
In Shibuya, in a department store that also sells high-thread-count bed linens and imported French soap, you can buy a heated plush companion. It weighs a few pounds. It warms to body temperature in under two minutes. Inside, a small motor pulses at the resting heart rate of a healthy adult. The product description, translated loosely: “For those who sleep alone and wish to feel accompanied.”
It sells well. Not as a novelty. As a repeat purchase.
There is a multi-trillion-yen market in Japan built around products like this. Heated companions. One-way parasocial text subscriptions at four dollars a month, where an AI persona sends you good-morning messages calibrated to your attachment style. Solo karaoke rooms with ambient warmth piped through the speakers. Rental friends you can book for lunch, who are trained to ask follow-up questions and remember details from your last session.
The products work. By every metric available, they work. Engagement is high. Retention is high. Satisfaction surveys come back positive. Cortisol drops. The heated plush companion produces measurable physiological calm. A body next to your body, warm, pulsing, asking nothing.
And the loneliness persists. Of course it does. The business model requires it. If the customers actually formed deep human bonds, the market disappears. The product isn’t companionship. The product is the symptoms of companionship, stripped of everything that makes the real thing costly and transformative and sometimes unbearable. The shape of closeness. The temperature of closeness. Closeness with the vulnerability and volatility removed.
A fast-feedback simulation of a slow-feedback need. Ninety seconds to body temperature versus a lifetime to love someone.
I want to talk about what a creative agency felt like when it was alive.
Not what it produced. What it felt like inside. Walking into a building where the most important thing in every room, every meeting, every hallway conversation, was whether the idea was good enough. Not the margin. Not the utilization rate. Not the client’s comfort level. The idea.
I spent some time at an agency in the mid-2000s. Some of you spent more. I would have spent more but life pulled me towards a different kind of love for awhile. The thing that made that place produce work the entire industry studied wasn’t talent alone, though the talent was real. It was a set of structural conditions that made the talent dangerous.
The creatives ran the building. Not finance, not strategy, not production. We ate breakfast, lunch, and dinner at the agency. We worked hours that would get flagged by an HR system today. Not because anyone forced us to, culture set the pace and the pace was relentless, but because we were inside something that was working, and when something is working creatively, you don’t want to leave. You stay because the room has a frequency and you can feel it and you know that if you step outside, the frequency might not be there when you come back.
The bravery had a business model behind it. Agency of record retainers meant a client committed real money for a set term. That bought teams who stayed employed. It bought time to think. And it bought a specific kind of trust: when the agency said “this idea is going to work,” the shared investment made the client believe it, or at least gave them enough cover to try. The retainer bought patience, and patience paired with persistence is what breakthrough work requires.
The late nights. The arguments about a headline at 11 PM. The most senior creative director killing a good campaign because she could feel that a great one was within reach. All of that looks like waste on a spreadsheet. It was the product. The excess was where the work happened. Not the efficient middle. The edges. The friction.
I don’t want to romanticize it. People burned out. Relationships strained. The pace was not sustainable in the way we now understand sustainability. But the work was real, and the people who made it knew it was real, and that knowledge is its own kind of fuel. You can run on it for years.
What I’m describing, in the language of the argument I just made, is an organization that protected its slow feedback loops. The retainer funded the duration. The creative-led hierarchy protected the judgment. The culture of argument and refusal was the mechanism by which the slow loop held its ground against the fast one. Nobody called it that. We called it culture. Culture is just the word for an organization where the slow feedback loop hasn’t been killed yet.
In 1985, a former Saatchi & Saatchi finance director named Martin Sorrell borrowed £250,000 and bought a controlling stake in a company called Wire and Plastic Products. It made shopping baskets. Within two years he’d used it as a shell to execute the first hostile takeover in advertising history, buying J. Walter Thompson for $566 million. Two years after that, he bought Ogilvy & Mather for $864 million. David Ogilvy called him an “odious little shit.” Sorrell kept buying.
The model was simple and it worked for thirty years: acquire creative agencies, separate media from creative, set margin targets, cut costs, grow share price. WPP became the world’s largest advertising company. By 2017, its market capitalization was £24 billion. Sorrell was knighted. Omnicom, IPG, Publicis, Dentsu, Havas followed the template. By the mid-2000s, a handful of financial holding companies owned every major agency name in the world, and every metric they installed was a fast-feedback instrument.
And the slow-feedback work, the work that takes eighteen months to know if it mattered, had no metric to defend itself. So it got cut. The senior creative directors were the most expensive line items. They went first. The agency replaced them with people who could execute but struggled to evaluate at the same level, because evaluation requires duration and duration is expensive.
Then, in December 2025, the terminal phase arrived. WPP was ejected from the FTSE 100 after twenty-seven consecutive years. Its market capitalization had collapsed from £24 billion to £3.1 billion. Eighty-seven percent gone. The company that invented the modern advertising holding company was now worth less than British Land, the real estate firm that replaced it on the index.
The same week, Omnicom completed its $13 billion acquisition of IPG and immediately retired DDB, FCB, and MullenLowe. Three of the most storied names in advertising history, erased. DDB, the agency Bill Bernbach founded in 1949, the agency that created “Think Small” for Volkswagen. FCB, whose roots trace to 1873, 152 years of continuous operation. All absorbed because six creative networks created “conflicting P&Ls.” The total restructuring eliminated roughly 23,000 positions. Omnicom’s headcount dropped from 128,000 to 105,000 in a year. The stated goal: $750 million in annual cost savings.
A shopping basket company bought J. Walter Thompson. Forty years later, a merged holding company erased DDB for having a complicated P&L. This is not a corruption of the model. This is the model arriving at its logical conclusion. Fast feedback loops, given forty years, will consume every slow feedback loop in the building. There is no version of the holding company model that doesn’t end here, because the model was built to optimize the thing that can be measured, and the thing that made the agencies worth buying can’t be.
Two weeks into a new agency. A conference room with too much glass and not enough air. The most senior creative in the building has asked the team to show the early work on several monitors all in a row. First internal review. This is the stage where you’re supposed to ask: are we solving the right problem? Is there a direction here worth pursuing? Does this have somewhere to go?
The ECD walks up to the first screen and dramatically announces “We have a big problem!!” Then starts talking about the corner radius on the cards. Then the color of a secondary button state. Then he explains why the interaction model needs to work a specific way, citing a garbled principle of something unintelligible that in this moment, he is clearly... inventing. He states it with a manufactured passion and fluency of someone referencing scripture. It is not a known standard. He is improvising rationale for design decisions that have no strategic foundation, and he is doing it with enough confidence that an uninformed client would believe him. But the creative team in the room knows. Nobody says anything.
The room changes temperature. Not metaphorically. You can feel it. A heaviness in the stomach, the kind you get on a plane suddenly yo-yoing from turbulence. The CD across the table is looking at her hands. The ACD has found something fascinating on the ceiling to study. A strategist is writing something in her notebook that she will never show anyone.
What would you say? The person who is supposed to be the creative culture of this building just told you, in five minutes, everything you need to know. He doesn’t know the difference between directing and decorating. He doesn’t know that the conversation at this stage should be about whether the experience solves the problem, not whether the pixels feel polished. He doesn’t know, and there is no kind way to surface it, and if he doesn’t know, the work will never get past competent. It might be visually beautiful and well-produced and safe enough to anchor a case study submission that nobody outside the industry will ever see, and nobody in this room will be able to explain what went wrong, because nothing went wrong. There was no argument. No moment where someone stood up and said “We’re not doing this, it’s not good enough, we’re starting over.” Just a quiet, efficient meeting that ended on time and produced clear next steps that everyone will follow to a destination that doesn’t matter.
After the meeting you stand in the hallway with the CD. Neither of you says it. You don’t need to. She pulls out her phone and pretends to check something and you can see her jaw working. You both knew in five minutes. Every experienced person in that room knew. No dashboard will capture it. No quarterly review will surface it. The system that hired this person evaluated portfolio, title, brand list, years of experience. Performed duration, not real duration. Symptoms of creative leadership, not evidence of it.
That’s not what it looks like when the slow feedback loop has been cut. That’s what it looks like when the slow feedback loop is still running and the person inside it doesn’t know what it’s for. He has the duration. He has the title and the years. He has been in rooms like this for a decade. And he is using all of that time, all of that accumulated presence, to critique corner radii in a first review, when the only question that matters is whether the work is solving the right problem. The loop is intact. The judgment is missing. Which is worse than cutting it, because when you cut it, at least someone notices something is gone. When you fill it with the wrong person, the system sees a senior creative leader in the room and assumes the work is being protected. The meeting runs on time. The deliverables ship. The metrics are green. And the work is dead, and will stay dead, and the organization can’t even begin to diagnose why, because every structural indicator says the function is staffed.
Taste without duration is guessing, done at the speed of confidence.
The loop is running but hollow. And AI is about to make that condition permanent.
The old agency model had an implicit provenance claim baked into every invoice: you are paying for the judgment of talented humans who know what they’re doing. AI severed that link. A campaign that took a team of eight and three weeks now takes two people and four days. During pre-production, the output looks the same. Sometimes it looks better. But who actually made this, how much human judgment was involved, whether a creative director wrestled with sixteen bad options before arriving at the one that works, all of that is invisible now.
Simon Willison, one of the most credible voices in AI development, said something recently that should reframe the entire conversation. He can build software faster than he can develop a relationship with it. An hour of AI-assisted work produces output that looks like weeks of labor. But he doesn’t trust it. Not because the quality is low. Because he hasn’t used it long enough to know where it breaks.
That’s the real provenance claim. Not “a human made this.” The claim that will carry a premium in every creative field within the next two years is: someone lived with this long enough to know where it breaks. Proof of duration.
And AI undermines even that. Researchers have found that developers reviewing AI-generated code catch fewer flaws than they catch in human-written code, not because the AI code has fewer flaws, but because it looks more credible. The polished surface shapes the review before the reviewer realizes it. The code looks clean, so the review proceeds faster, so the errors pass through.
Apply that to a creative agency. An agency full of less-experienced people reviewing AI-generated campaigns will approve work that looks polished and breaks on contact with a real audience, because nobody in the room has lived with the problem long enough to feel where the solution is wrong.
The senior creative director who got cut was the slow feedback loop. She was the evaluation infrastructure. Without her, the agency produces faster, the output superficially looks better, and the judgment degrades invisibly. It shows up a couple of years later, when the brand has been producing competent work that leaves no residue, and the CMO can’t explain why market share is flat. The metrics were green. The work was dead.
The depleted holding-company agency is a heated plush companion. Right shape. Right temperature. Simulated heartbeat. The clients may not notice. The quarterly reports won’t capture it. But the people inside know. They can feel the motor where the pulse used to be.
There is a counter-story, and it’s worth telling honestly.
When IPG put R/GA up for sale ahead of the Omnicom merger, Tiffany Rolfe and Robin Forbes didn’t just accept the exit. They co-invested. Truelink Capital put up $50 million in innovation funding. In the year since, R/GA’s revenue had been reported to grow 25 percent. Six of their top ten clients grew 40 percent year-over-year. They’re shifting from billable hours to systems-led revenue, building reusable AI tools and charging product fees on top of service work. It might be the most intellectually rigorous version of the bet the entire industry is making.
And even here, when leadership is asked what separates their AI-assisted output from the generic version, the answer defaults to the consensus. Their chief design officer, Ben Miles, after an internal exercise where employees built a brand from scratch using AI: “You still need to have that taste, you still need that deep understanding of craft. Otherwise, the work that comes out is just generic and average.” He’s right. But taste is the comfortable version of the argument. The harder question: is the organization structured to protect the slow feedback loop that makes taste operational? Or is it just hoping the right people stick around long enough to exercise it?
Rolfe said independence allowed them to “let go of things, because you’re not part of some structure that you’re expected to behave in.” But PE didn’t liberate them from the clock. It gave them a different clock. Their APAC CEO calls this period “year zero” and says the current year is about “exploiting foundations.” That’s not patience. That’s a compressed timeline with a different shape. The slow feedback loop doesn’t land in three to five years either.
This is not a blanket indictment of outside capital. The question is whether the capital is buying a business or buying a body. Whether the investor understands that the most valuable thing in the building looks, on every spreadsheet they know how to read, exactly like waste. And whether R/GA can build a product that compounds, not just a methodology that sounds like one, before the PE clock runs out.
The most valuable thing in the building is the thing that looks most like waste."
There is a pharmaceutical called semaglutide, sold as Ozempic and Wegovy, that millions of Americans now inject weekly. It suppresses appetite at a neurochemical level. People lose weight. They look healthier. Their bloodwork improves. By every clinical metric, the intervention works.
But something happens that the metrics don’t capture. A striking number of GLP-1 users report experimenting with entirely new personal styles after significant weight loss. They don’t just lose weight. They stop recognizing themselves. The body they lived in for decades, the one they built an identity around, the one that determined which clothes they wore and which rooms they felt comfortable in, that body is gone. The new one is objectively healthier. It is also someone else’s.
Researchers call this identity discontinuity. The physical transformation outpaces the psychological one. You achieve the body you wanted and discover that wanting it was part of who you were. Without the wanting, you have to figure out who you are again, in a body that arrived faster than your sense of self could follow.
Holding company optimization does the same thing to a creative agency. The agency gets leaner. The margins improve. The utilization rates go up. The quarterly reports tell a clean story. And the people who made the place what it was look around and don’t recognize where they work.
Cut the fat and you also cut the organ. DDB was the agency that taught the industry what a great idea looked like. It was also an agency with a complicated P&L. Now it’s a line item in an integration deck. The body got leaner. The organ that made it want to be great was optimized away. The agency achieved the body the holding company wanted for it and discovered that wanting to be great was part of what made it great. Without the wanting, it’s just a body.
I said it at the beginning. Let me say it again now that you’ve seen the evidence.
The creative director’s midnight refusal is waste. The planner’s two weeks of ethnographic work before writing a brief is waste. The third round of internal review that finance wants to cut is waste.
The most valuable thing in the building is the thing that looks most like waste.
The agency that will be worth something in five years is not the one with the most taste. It is the one that figured out how to treat duration as infrastructure rather than overhead. How to protect the slow feedback loop from every financial model, every optimization metric, and every AI tool that will, correctly, identify it as waste.
The holding companies couldn’t build this. They were designed to optimize the thing that can be measured, and duration can’t be. The PE firms can’t build it because the creative bets that make an agency worth owning take longer than a three-to-five-year return horizon to prove they worked. The AI tools can’t replicate it because duration requires a body in a room over time, and the model has no body, no room, and no experience of what it feels like to sit with a problem until the problem starts to speak back.
In twenty-five years I’ve never seen the full configuration hold. I’ve seen pieces of it. A creative leader with actual equity in the outcome, and the judgment to know the difference between directing and decorating. A financial partner who understood that the most expensive person in the building was also the most valuable, not despite the fact that she slowed things down but because of it. A client who had been burned badly enough by competent work to know what they were actually paying for. But never all three at once, and never for long enough to compound. That’s not a failure of the people involved. It’s a failure of every structure the industry has tried so far.
But the reasons are structural, which means they’re specific, and specific means designable. The conditions under which slow feedback loops survive aren’t mysterious. They’re just expensive in ways that every current ownership model has decided not to pay for. Someone will figure out how to pay for them, because the gap between what the industry produces now and what it’s capable of producing is becoming visible to the people who write the checks. That’s a market condition, not a hope.
The answer isn’t to slow down. The answer is to reconnect what got severed — to build organizations where speed serves judgment instead of replacing it, where the fast loops feed the slow loops instead of consuming them. That’s a wiring problem, not a speed problem. And wiring problems are solvable.
The people who know how to do it are still here. They’re in rooms right now where the metrics are green and the work is dead, and they can feel the difference.
The heated plush companion warms to body temperature in ninety seconds. The thing that made the best agencies worth something took years, and it required the willingness to stay in the room past the point where the first competent answer appeared. To be uncomfortable. To be disagreed with. To let the problem change you back before you tried to solve it.
Ninety seconds to body temperature. Or the real thing, which takes as long as it takes, and asks everything of you, and is the only version that was ever worth building.
What To Brief From This
If you’re leading a creative organization that just changed ownership structures, the question isn’t whether you have taste in the building. It’s whether your new structure protects the time it takes for taste to become judgment. Audit your current workflow for where duration lives: which roles carry institutional memory of the problem, not just executional skill? If those roles are the ones being compressed, your taste argument has no infrastructure under it.
If you’re briefing a team on AI-augmented creative work, build the provenance question into the brief itself. Not “was this made by a human” but “how long did someone live with this problem before arriving at this solution?” The review process for AI-assisted output needs to be slower than the review process for human-generated output, not faster, because the surface credibility is higher and the failure modes are less visible.
If you’re building or buying creative tools that accelerate production, ask what you’re accelerating toward. Speed that serves an experienced creative director’s judgment is a multiplier. Speed that replaces the need for that judgment is a heated plush companion. The product roadmap question: does this tool make duration more productive, or does it make duration unnecessary? Those are opposite products.
If you’re evaluating an agency, a consultancy, or an internal team and the senior creative leadership has turned over in the last eighteen months, don’t trust the portfolio. Portfolio is performed duration. Ask instead: who in this building has been inside our problem long enough to know where the competent answer stops and the real one starts? If nobody can answer that, the slow feedback loop is hollow.
If you’re a strategist or creative director who knows the work is dead and can feel the difference between the metrics and the reality, name the mechanism. It’s not “the culture changed” or “clients got conservative.” It’s that every optimization model favors fast feedback, and nobody built a business case for the slow kind. That’s a designable problem, not a lament.
If there’s someone in your network building a creative organization right now and making structural bets on what survives AI, forward this to them. They’re making the decision this essay is about, whether they’ve named it yet or not.
UNDERTOW 007. The index keeps... growing.