Meta Bet $145 Billion on AI Agents. Zuckerberg Just Told Staff They Haven't Kept Up.

On July 2, at an internal town hall, Mark Zuckerberg told Meta staff that the technology he has bet the company on is running behind. Meta is spending roughly $145 billion on AI infrastructure this year and reorganized itself around it. The CEO’s own read, in a recording heard by Reuters, was blunt:

“The trajectory of the agentic development over at least the last four months hasn’t really accelerated in the way that we expected.”

When the company spending $145 billion says the agents aren’t keeping pace, the marketing team spending a few hundred thousand should recalculate too.

Zuckerberg wasn’t hedging for the cameras. He was explaining a reorganization to the people living inside it. Meta laid off about 10% of its workforce in May and moved roughly 7,000 employees into AI-focused teams, a bet he told staff wasn’t as “clean” as intended and hadn’t “come to fruition yet.” He said executives had been “super optimistic” about tools like Anthropic’s Claude Code back in January and February, and now expects meaningful benefits only “within the next three to six months.” A $145 billion AI program just put a two-quarter asterisk on its own timeline.

Marketing is making the same bet with less room to be wrong

That should land hard on marketing, because the promise finance keeps hearing is the same one Zuckerberg just discounted. CMOs are putting 15.3% of their budgets into AI while only about a third say they’re ready to scale it. Vendors are shipping “end-to-end” agents that pitch running the marketing loop on their own. And the results so far rhyme with Meta’s: research on marketing leaders found just 12% of AI pilots have reached a fully integrated, productive stage.

The gap has less to do with the technology’s ambition than with the readiness of the organization using it. Ralf Strauss, chairman of the European Marketing Confederation, named the bottleneck in that same research:

“While technology develops exponentially, organisations remain pretty flat in their development. The bottleneck is in the majority of the companies know-how.”

That’s the through-line from Meta’s town hall to a 30-person marketing team’s Q3 plan. Buying the agent is the easy part. Building the process, the oversight, and the skill to run it is the slow part, and no amount of infrastructure spend accelerates it on command.

The honest plan builds in the wait

None of this says agents are a dead end. Meta clearly believes the payoff is coming, just later than the slide promised. The lesson for marketing is about sequencing and telling the truth about timelines. Gartner, cited in EY’s June analysis, expects 40% of agentic AI projects to be canceled by the end of 2027 on cost and unclear value. The projects that survive will be the ones whose owners planned for the slow part instead of the demo.

So do what Zuckerberg just did, only earlier and cheaper. Mark your agent timelines down by two quarters before finance marks them down for you. Fund the process work, the review loops and the skills, at least as heavily as the tools. And judge every agent pilot against a production bar, not a demo, because a pilot that never reaches the point where it returns real money is a cost with a story attached. If the company spending $145 billion is telling its own people to wait, the honest marketing plan builds in the wait.