Every few months someone announces that AI is about to "solve" marketing. Fully autonomous growth teams. A model that writes the ads, picks the channels, allocates the budget, and prints customers while you sleep. The pitch always carries the same implicit promise: marketing is a problem, problems get solved, and the solver is almost here.
I think this is wrong, and not because the models are too weak. It is wrong because of what kind of game marketing is.
The thesis
Marketing cannot be solved the way chess was, because it is not a stationary problem. It is an adversarial arbitrage market where every discovered edge destroys itself. AI will industrialize the execution—and the only structural consequence is that the alpha decays faster.
There will be AI everywhere in marketing. There will be no AGI moment for marketing—no point where the game is finished and the best model simply wins from then on.
Marketing is an attention market
Strip away the creative mystique and marketing is a pricing game: how many dollars buy how much attention, and how much of that attention converts.
The mechanics make this literal. Nearly every distribution channel that matters—search, social, video, app stores—clears through an auction. Prices are not set by anyone; they emerge from how many buyers want the same eyeballs at the same moment. When a channel is crowded, CPMs rise until the marginal advertiser drops out. When it empties, prices fall until someone notices the bargain.
Run that process long enough and you get the marketing version of the efficient market hypothesis: customer acquisition cost gets pushed toward customer lifetime value, across the whole market, for everyone at once. If a channel reliably returns three dollars for one, money floods in until it doesn't.
Which means sustained outperformance in marketing has exactly one source—the same source it has in trading. You have to find a mispricing before the crowd does.
Three arbitrages I have watched
The pattern is easiest to see in specific trades.
ChatGPT wrappers, 2023. When ChatGPT detonated, consumer curiosity about AI exploded years ahead of the ad market's ability to price it. Keywords and creatives touching AI converted absurdly well while still costing what they had cost the year before. A wave of thin wrapper apps made real money in that window—not because the products were good, but because attention around AI was briefly, wildly underpriced. Then every app on earth added "AI" to its name, auctions repriced, and the window shut.
Workday on YouTube, 2020. When the pandemic hit, brand advertisers panicked and pulled budgets. YouTube inventory went on clearance. Most companies saw uncertainty; a few saw the cheapest reach in a decade and bought aggressively while their competitors went dark. That was not a creative insight. It was a liquidity trade—buying attention at panic prices from forced sellers.
Base44, 2025. Same trade, new decade: heavy spend into a channel the competition had quietly abandoned as unfashionable, at prices that reflected the absence of competition rather than the absence of customers.
Three different eras, one identical shape: spot the mispricing, size the position, ride it until the crowd arrives, get out or get diluted. Nobody in these stories "solved marketing." They noticed a price that was temporarily wrong.
Every edge self-destructs
Here is the uncomfortable property that follows: in marketing, the act of winning publishes the strategy.
Your ads run in public. Your landing pages are one click away. Your growth loops get reverse-engineered in Slack channels and podcast interviews. Ad libraries make competitive espionage a product feature. Whatever worked for you this quarter is, by definition, visible to everyone it worked on.
So every edge follows the same decay curve. A meme format mints attention for two weeks, then becomes the surest signal of a brand trying too hard. A cold-email template gets 30% reply rates until the ten-thousandth send trains everyone's filters. A channel prints customers until the case study circulates, and then the case study itself raises the price. Quants call this alpha decay: publish the signal, kill the signal. Marketing runs the same physics with a shorter half-life, because in marketing, shipping the signal is publishing it.
This is why marketing playbooks age like fish while engineering playbooks age like wine. A database indexing technique from 2005 still works, because the database does not adapt to being understood. An acquisition tactic from 2023 is already dead, because the market it worked on absorbed it.
Where the quant analogy breaks
I want to be honest about the limits of the comparison, because they sharpen the thesis rather than weaken it.
First, marketing is not strictly zero-sum. Advertising can create demand, not merely reallocate it—categories get built, and the attention pool itself grows as new platforms emerge. The zero-sum crunch is local: at any given moment, in any given auction, one impression goes to one bidder. The game is zero-sum at the margin even while the pie grows in the aggregate.
Second, marketing's prices are far noisier than financial markets'. Attribution is murky, feedback loops take weeks, and most participants are not even trying to be rational. This means inefficiencies persist longer than they would on a trading floor—which is precisely why the arbitrage game is playable at all. Marketing is not an efficient market. It is an inefficient market perpetually in the process of becoming efficient, one discovered edge at a time.
And third: brand sits partially outside the game. Distinctive memory structures—what people recall unprompted when the category comes up—decay on a timescale of years, not weeks. Brand is the one marketing asset that behaves less like a trading signal and more like compounding equity. It is not exempt from competition, but it is exempt from the two-week half-life.
What "solved" would even mean
Now the core argument. Chess was solvable-in-practice because chess is stationary: the rules never change, and the board does not study you back. Protein folding, same shape. These are problems where the environment holds still while you get smarter, so intelligence accumulates into a permanent advantage.
Marketing's environment is other optimizers. The thing you are optimizing against is not a fixed landscape of consumer behavior—it is thousands of other marketers reading the same dashboards, plus an audience whose taste actively mutates in response to being targeted. The moment any strategy becomes legible, the environment reprices it. You are not solving a puzzle. You are playing poker against people who can see your discarded hands.
Run the thought experiment: suppose someone actually built the marketing AGI—a system that reliably finds underpriced attention. What happens next is not victory. What happens next is distribution. Everyone runs it, every mispricing it can see gets bid away the moment it appears, and the system's edge converges to zero because it works. A solver everyone owns is a solver that works for no one.
That is the deep difference. Chess engines got better and stayed better. A marketing engine, at the limit, deletes the inefficiencies it feeds on. The terminal state is not "marketing is solved." The terminal state is a faster market with thinner, shorter-lived spreads—and the game continues at a higher metabolism.
What AI actually changes
None of this means AI is irrelevant to marketing. It means AI's role is specific: it collapses the cost of execution, and therefore compresses the arbitrage cycle.
Creative production used to take an agency two weeks; now it takes an afternoon. Testing ten positioning angles used to be a quarter's roadmap; now it is a batch job. The loop of notice-mispricing, exploit, saturate used to take months. It is heading toward days.
Notice what that does and does not change. It lowers the floor—baseline execution becomes commodity-grade for everyone, so incompetence stops being the differentiator. It raises the tempo—edges get found faster and die faster. What it cannot do is manufacture durable edge, because every capability it grants you, it grants your competitor in the same app store. AI in marketing is a faster car in a race where everyone gets the same car. The lap times drop. The finishing order still depends on picking lines nobody else saw.
If anything, the endgame favors the arbitrage view even more: when execution is free, the only scarce inputs left are earliness and judgment about where the mispricing is.
So what
If marketing is arbitrage rather than a solvable problem, a few practical things follow.
- Earliness beats optimization. Being mediocre in an empty channel outperforms being excellent in a crowded one. The Workday trade required no genius creative—it required showing up when others fled.
- Expect decay; plan for it. Any tactic that works should come with an expiration mindset. The question is never "is this working?" but "how long until this stops working, and what is next?"
- Speed is the only durable meta-strategy. You cannot own an edge, but you can own the machinery that finds the next one faster than the market reprices the current one.
- Put compounding where compounding works. Tactics decay; brand and product accrue. Rent attention through arbitrage, but convert it into the assets that do not reprice overnight.
And treat anyone selling you a permanent growth playbook the way you would treat someone selling a permanent trading strategy: if it truly worked forever, it would not be for sale.
Closing thought
Chess had an AGI moment because the board never fights back. In marketing, the board is the other players—so intelligence does not end the game; it just raises the speed at which the game is played. There is no finish line here, only a faster track. The prize does not go to whoever solves marketing. It goes, over and over, to whoever notices what everyone else is about to notice—first.