Three Marketers and an Infinite Compute Budget

A non-zero number of corporate marketing departments would produce way better output if they let 80% of the org go and handed the top two or three people an effectively infinite compute budget. And I mean that pretty much literally: the CMO keeps the highest-agency people, prunes the rest of the org chart, and reallocates all that saved headcount budget directly into API tokens and tools.

That might not apply to every single company or every flavor of marketing. But when you actually audit how a typical 30-person marketing org spends its weekly cycles - mostly coordination overhead, status meetings, and passing drafts up and down an approval chain - and compare that to what one cracked generalist with taste and a fleet of frontier models can ship in an afternoon, the leverage gap is just massive.

We’ve crossed the threshold where leaders should be looking at that math as a serious architectural refactor, rather than an abstract // spicy thought experiment.

What the department ships

If you profile how a mid-sized marketing org actually spends its cycles, the workload basically splits into two buckets.

Bucket 1 is actual artifact generation: landing page copy, email sequences, ad variants, social posts, blog articles, sales one-pagers, slide decks, lightweight graphics, and campaign readouts. Under the hood, this is all just structured language and a bit of visual layout that gets packaged up and shipped.

Bucket 2 is pure inter-node communication overhead: writing creative briefs, running multi-stage review loops, navigating approval chains, sitting in status syncs, the classic Monday standup where everyone verbally serializes the exact updates they already posted in Slack, and writing meta-docs to explain other docs.

Two years ago, the physics of human bandwidth meant you had to scale up headcount just to brute-force Bucket 1 - and you paid for that parallelization by eating the massive O(N^2) coordination tax of Bucket 2.

Today, one high-taste marketer wired into frontier models can zero-shot an 80%-there first draft of almost every single artifact in Bucket 1 in a couple of minutes. And here’s the kicker: once you no longer need a large cluster of humans to execute Bucket 1, the entire coordination apparatus of Bucket 2 just evaporates. You don't just save on the production work—you get to garbage-collect the bureaucracy that only existed to synchronize the people doing the production work.

The team as a distributed system

It’s clarifying to look at a marketing org the exact same way a systems engineer looks at a distributed cluster. Every human on the team is a compute node. Every time you add a node, you get a linear bump in raw compute, but you take a quadratic hit on inter-node communication overhead—literally N(N-1)/2. A 10-person team has 45 communication channels to keep in sync; a 20-person team has 190. In software engineering, we learned decades ago (Brooks’s Law) that past a certain cluster size, the system spends way more cycles passing state updates around than doing actual forward computation.

Marketing teams hit that scaling wall even faster than engineering teams. Because their primary output artifact is natural language, customers immediately feel it when outputs generated by twenty different people don’t share a single coherent voice. So the org has to invent a massive reconciliation layer just to enforce consistency: a brand lead redlines the blog post, a PMM scrubs the technical claims in the sales one-pager, and the CMO ends up completely rewriting the landing page hero copy at 11 p.m. In almost every mid-sized team I’ve looked at, the highest-leverage people spend 80% of their week acting as human git merge resolvers - manually patching other people’s diffs just so the company sounds like one brain.

When you collapse that cluster down to three high-taste generalists, almost all of that message-passing overhead drops to zero. Three people can keep the entire brand context window loaded in their heads at once. They don’t need to write a 4-page creative brief just to serialize what the customer cares about to someone else. They can hop on a 5-minute huddle and resolve a state conflict that a 20-person org would bounce around in Google Docs review loops for two weeks.

Where the tokens go

People hear "AI budget" and picture a ChatGPT seat per employee. I mean something much bigger than that: a few people who run models all day, on everything, with no one telling them to economize.

In practice, a strong marketer with a big budget might:

• feed every sales call transcript from the last year into a model and pull out the exact phrases customers use to describe their problem
• generate fifty versions of a landing page headline, cut them to five by hand, and test those five
• build and ship a landing page themselves instead of filing a ticket with the web team
• have agents draft the whole email sequence, the ad set, and the social posts for a launch in one afternoon, then spend the rest of the week editing
• point an agent at campaign data every morning and read a summary instead of waiting for the monthly report
• research a new market segment in a day instead of commissioning a study

A heavy user can spend thousands of dollars a month on tokens this way and still cost the company less than one junior hire. The company also skips the management overhead, the onboarding, and the coordination tax it pays for every new person.

Generation is cheap; verification is the bottleneck

When a marketer generates fifty headlines, someone has to know which three are good. When a marketer has an agent draft a launch sequence, someone has to notice that email four makes a claim legal will hate, that the tone drifts in the middle, that the whole stack answers a question no customer asked. Marketers move from writing to reviewing; they review well only if they have taste, real knowledge of the customer, and a fast, reliable sense of what's off.

Your best performers have this. That's usually why they're your best performers. They can read a draft and know in ten seconds whether it works; they can also say why. Give them a model and they become something like a small studio. They generate a lot, throw most of it away, and ship the good part.

Weaker performers often don't have it yet. Give them the same model and they generate a lot, keep most of it, and ship it. You get more content at roughly the same quality, which in marketing usually means more noise in front of the customer. I'd argue a lot of teams are in this state right now: everyone has AI access, the team ships more than ever, and nobody can point to better results.

Average work lost its price

Before these tools, junior marketers did the high-volume, lower-stakes work, and companies paid for that volume. A first draft of a blog post had real value because someone had to spend hours writing it.

Now anyone can get a draft of that blog post in thirty seconds. A junior marketer would have written something of about the same quality, sometimes better, sometimes worse in ways only an experienced reader would catch. Companies used to pay a salary for that draft; now they pay almost nothing.

So companies should pay very little for average marketing output; they should pay a lot for the judgment that sits on top of it.

None of this says junior marketers are bad at their jobs. Most of them do exactly what companies hired them to do. Companies just no longer need as much of it from people.

Where this doesn’t work

Relationship work. Some marketers spend their week talking to people: running events, managing partners, handling analysts and press, sitting in on customer calls. Nobody can hand that to a model. If you cut those people, you inevitably lose something you can't buy back with tokens.

Picking the best performers. Half of the companies I talk to can't tell who their best marketers are. They promote the loudest person, or whoever manages up well, or the asshole whose campaigns happened to run during a good quarter. If leaders keep the wrong three people, they get bad judgment at ten times the volume.

Apprenticeship. Today's best performers learned taste by writing a lot of mediocre copy and having someone better tear it apart. If companies stop hiring juniors, they stop growing seniors. I don't have a clean answer here. My guess is that the strongest teams will keep one or two juniors on purpose and treat them as apprentices. The juniors review model output alongside a senior, learn why the senior cuts what they cut, and build taste faster than people could when they had to write every draft by hand.

Sameness. If every company's three marketers use the same models, prompt them in similar ways, and edit toward the same idea of "good", customers start hearing one voice from every company. The best performers push against this by bringing in what they can't get from a model: specific customer stories, opinions the company actually holds, a voice someone built over years. Teams that skip this will save money and lose what made them distinct.

Timing. People still have to catch the mistakes agents make; some of those mistakes are subtle. I think leaders should size the team partly on how good the tools are this quarter, then revisit that as the tools change.

What I'd actually do

If you’re running a marketing org today, step zero is to run a super simple dependency audit. Pull up every single artifact your team shipped last quarter, and right next to each item, write down the name of the actual human who supplied the taste - whoever looked at the thing and decided what "good" looked like. In almost every org, that second column collapses down to the exact same two or three names over and over.

Those are the nodes you keep.

From there, give that tiny core team an API and tooling budget large enough that they never once have to think about token costs, tell them to automate the entire production pipeline end-to-end, and evaluate them strictly on two metrics: shipping velocity and actual pipeline impact. Keep the high-touch human relationship roles (like events or partner marketing) where being a real person in a room actually matters, and keep one or two super hungry juniors around as apprentices to learn the craft. Prune everything else - especially the middle-management coordination layer, because a 3-person cluster has zero need for a routing layer.

Expect a temporary dip in throughput during Q1 while the team tears down the legacy workflows and builds out their new automation harness, and by the end of Q2, you’ll have a clean, empirical read on whether the bet worked.

Anyway, that’s roughly where my mental model has landed on this. If you’re a CMO or marketing lead who has actually run some version of this experiment in the wild, I’m super curious to hear how it played out - if it blew up in practice…