Somewhere in the last few months, "we replaced our marketing team with AI" became a headline instead of a warning. ClickUp said it was running with one CMO and three thousand agents instead of fifty marketers. Meta tried a version of the same idea internally, tracking employee keystrokes to see whether agents could do the job instead. One of those stories held up. The other one got walked back within months, quietly, after it stopped working.
If you're staring at your own marketing spend right now wondering whether the fix is fewer people and more AI, it's worth knowing which of those two stories you're actually looking at.
Meta's plan, reported by Reuters as "Project OT," put small pods of AI agents in charge of work that used to belong to real employees. To see whether it was working, the company tracked keystrokes. Employees pushed back. The agents underperformed. Within months, Zuckerberg walked the whole thing back.
That's not a story about AI failing. It's a story about what happens when you try to replace a system with a shortcut. The people doing the work weren't the bottleneck. The plan was.
We had a client at $5M ARR with a marketing team that was already good at their jobs. Smart people, clear roles, real output. They still hit a wall. When we looked at where the bandwidth was going, the problem wasn't headcount. It was that the team was falling behind on which AI tools were actually worth adopting, inside a plan that already had clear direction.
Same lesson, two very different rooms. AI is leverage on a system that already works. It is not a replacement for the system.
Most companies looking at their marketing budget right now are asking the wrong question. Not "how many people can AI replace," but "does our marketing function operate as a system with a clear plan, or as a set of activities someone has to keep running." A system gets stronger with AI built into it. A pile of disconnected activities just gets faster at being disconnected — and faster isn't the same as working.
Here's a short way to check which one you have. Look at what your team actually does in a given week, and sort it into two piles. The first pile is execution: a draft, a variant, a report, research pulled together for a meeting. AI absorbs almost all of that today, and it should. The second pile is judgment: what to prioritize, what to say to the market, who to target, what to cut. That pile doesn't shrink because a tool got smarter. It still needs someone who's done this before.
If most of your team's time is in the first pile, you don't have a headcount problem, you have a tooling problem, and it's a fast fix. If the second pile is thin or nobody owns it, that's the actual reason growth has stalled, and no number of agents closes that gap.
Before you decide whether to cut headcount or buy another AI subscription, get clear on which problem you actually have. That's the first thing we work through in a Growth Workshop — not a pitch for a bigger team, a real look at where the plan is thin and where the execution is.
If the honest answer right now is "I'm not sure whether we have a people problem or a plan problem," that's exactly what the workshop is built to find out. Thirty minutes, a real diagnosis, no pitch.
This post pulled from Episode 1 of the Growth Unfiltered Show, "3,000 AI Agents vs. 50 Marketers: Who Actually Wins?" Antoine Vial sits down with Brian Graf and Welcome Read for the full debate on ClickUp's restructuring, what belongs to AI and what doesn't, and the rest of that week's GTM and AI news. Watch the full episode below.