I run a marketing firm, so understand that everything in this paper is me describing the asteroid from the dinosaur's side of the field. X Network bills clients for strategy, creative, media, and execution, the classic bundle, and I can tell you exactly which of those line items my own AI tooling has already hollowed out. When the person selling you the service tells you which parts of it are becoming worthless, it is worth listening, because I have every financial incentive to say otherwise.
The agency model has always been a bundle held together by client ignorance in the most neutral sense of the word: clients could not easily see which parts of the retainer were scarce judgment and which parts were replicable labor, so they bought the whole package. Media buying subsidized strategy. Production margins subsidized account management. The bundle worked because unbundling it required the client to do integration work they had no appetite for. AI just volunteered for that integration work, at near zero marginal cost, around the clock.
Here is my thesis, and I will make it falsifiable: by 2030, the traditional full service retainer will be a minority of agency industry revenue, displaced by three cleaner trades, judgment sold at a premium, machine execution sold near cost, and accountability sold as a service. Holding companies that survive will have shrunk headcount dramatically while holding revenue roughly flat, and the fastest growing firms in the industry will be under fifty people. If the bundled retainer is still the dominant revenue structure in 2031, I was wrong. I am not wrong, because I am already repricing my own firm accordingly.
Key Findings
- The agency bundle was an information asymmetry business. AI collapsed the asymmetry by making execution legible and cheap, and the bundle cannot survive legibility.
- Agency value splits into three separable products: judgment, execution, and accountability. Each has a different buyer, margin, and future.
- Execution is repricing toward machine cost regardless of anyone's feelings. Firms still billing human hours for machine work are harvesting a runway, not running a strategy.
- The scarce asset is what I call Taste Liability: judgment that a named human stands behind with reputation and contractual skin. Machines can generate options; they cannot yet be sued, fired, or embarrassed.
- Client side in housing accelerates, then partially reverses, because owning autonomous tooling turns out to be an operations burden most brands do not want.
- The winning firm of 2030 looks like a boutique: small senior bench, heavy proprietary tooling, outcome linked fees, and a reputation concentrated in identifiable people rather than a holding company brand.
What the Bundle Actually Contained
Unbundling analysis only works if you name the parts honestly. The classic retainer contained four things. Judgment: what should this brand do, which is rarer than the industry pretends. Execution: making the assets, trafficking the campaigns, pulling the reports, which is most of the hours billed. Access: relationships with platforms, publishers, and talent, plus preferential rates. And insurance: someone external to blame, a function nobody puts in the deck and everyone quietly buys.
For decades these were priced as one number, and the blend was profitable because clients could not benchmark the components separately. Now they can. A client can watch a machine produce forty ad variants in an afternoon and immediately re anchor what production is worth. They can watch an autonomous system reallocate budget, the shift I mapped in Paper No.50, and re anchor what campaign management is worth. Every component the machine touches becomes benchmarkable, and benchmarkable services converge to cost. The components the machine cannot touch, judgment and insurance, suddenly stand naked, which is uncomfortable, because agencies have historically been vague about how much of the fee they represented.
Taste Liability: The Product That Survives
Strip away everything replicable and what remains is a strange, durable product. Not ideas, machines have ideas by the thousand. Not analysis, machines analyze better than my best strategist on most dimensions. What remains is a human being with a track record saying: do this one, not those nine, and I stake my name on it. I call this Taste Liability, and I choose the word liability deliberately. The client is not buying the opinion. They are buying the fact that someone accountable holds it, someone who absorbs reputational and often financial damage when it is wrong.
Taste Liability explains dynamics that pure efficiency logic cannot. It explains why senior strategist rates are rising while junior rates collapse: liability does not delegate. It explains why the agency brand matters less and the individual practitioner brand matters more, a shift I have been telling clients to exploit for years in personal brand building, and which now applies with full force to the people selling marketing services themselves. A firm whose judgment is embodied in named, visible humans can charge for it. A firm whose judgment hides inside a logo cannot, because the client no longer believes the logo does the thinking, and the client is right.
The implication for agency economics is stark. Judgment does not scale with headcount, so the judgment business does not want headcount. It wants a small number of people whose word moves budgets, supported by tooling. That is not a smaller version of the old agency. It is a different species that happens to share a tax classification.
The Execution Layer Goes to Machine Cost
Meanwhile the execution business, the majority of industry hours, is repricing in one direction. Creative production, versioning, trafficking, reporting, optimization: every quarter the machine share of each grows and the defensible human premium shrinks. Firms respond in one of three ways. Some hide the shift, billing legacy hours for machine assisted work; this is margin harvesting and it ends the day procurement runs a benchmark, which is soon. Some pass the savings through and compete on volume; this is honest and brutal, and it turns execution into infrastructure with infrastructure margins. And some, the smart ones, productize: they wrap their execution stack in software, sell it as a platform with a service layer, and stop pricing in hours entirely.
Notice that all three responses end the pyramid. The old agency was an apprenticeship machine: juniors did execution, learned, and became seniors. When execution is machine work, the bottom of the pyramid has no economic reason to exist, and the industry has not begun to grapple with where its next generation of judgment will come from. The firms that solve deliberate apprenticeship, paying to train people the billing model no longer trains for free, are making a ten year investment their competitors will envy in exactly ten years.
The In Housing Whipsaw
Clients watching this unbundling draw an obvious conclusion: if execution is cheap machines and judgment is a few named humans, why pay an agency at all? So the in housing wave accelerates, and for eighteen months it looks brilliant. Then the second order costs arrive. The autonomous stack needs owners, mandate authors, auditors, vendor managers. The tooling landscape shifts quarterly and somebody has to keep up professionally. The named humans the brand hired for judgment discover that inside one company they see one company's data, while their agency counterparts see thirty clients' worth of pattern, and the judgment gap reopens.
So I expect a whipsaw: aggressive in housing through 2027, then a partial reversal into a hybrid where brands own their data, their mandates, and their platform relationships, and buy pattern recognition and accountability from outside. The agencies positioned for that reversal will be the ones that spent the interim becoming genuinely better at judgment across clients, and publishing the evidence. Visible, opinionated, testable thinking is the marketing that works for judgment businesses, which is why I practice what I preach about thought leadership and why these papers exist at all. You are reading my firm's product demo right now. I see no reason to pretend otherwise.
What I Would Do About It
If you run an agency: unbundle yourself before clients do it to you. Split your pricing into judgment, execution, and accountability this year, even if the numbers are uncomfortable, because the firm that reprices voluntarily keeps the client relationship and the firm that gets benchmarked loses it. Move execution to machine cost plus a thin margin, and put the margin you lost into the fees of your genuinely scarce people. Kill the pyramid gently: hire fewer juniors, but fund real apprenticeships for the ones you keep.
If you are client side: stop buying bundles, start buying components, and be honest about which component you actually lack. Most brands that think they need an agency need a mandate author and an auditor. Most brands that think they can fully in house are underestimating the pattern advantage of people who see many accounts. Contract for judgment with named individuals, not logos, and tie a real share of fees to outcomes.
And if you are a talented individual inside a big shop: your window is open. Taste Liability accrues to people, the tooling that used to require a holding company fits on a laptop, and the clients are learning to buy judgment directly. The great unbundling is not only happening to firms. It is the biggest talent liquidity event this industry has ever had. Act like it.