Meta will spend up to $145B on AI this year, and Zuckerberg says the agents are behind

Mark Zuckerberg admitted to Meta staff that its AI agents are behind schedule. In a July 2 town hall recorded and heard by Reuters, he said agent development over the past four months hasn’t “accelerated in the way we expected.”

That’s a rare public concession from the CEO betting the most on agents. Meta plans to spend as much as $145 billion on AI infrastructure this year, and Zuckerberg has bent the whole company around the bet: he cut about 10% of staff in May and moved roughly 7,000 people onto AI teams, including a group called Agent Transformation. Now he’s telling his own workforce the payoff isn’t here yet. He still expects it, only later. Bigger benefits should arrive in the next three to six months, he said.

What Zuckerberg told staff

The tone was unusually candid for an all-hands. Zuckerberg said the reorganization meant to speed agents up hasn’t paid off yet, and that the job cuts weren’t handled as cleanly as they should have been. Here’s what the reporting establishes:

• Agent development over the past four months hasn’t “accelerated in the way we expected,” Zuckerberg said, according to the recording Reuters reviewed.
• The gains from the May restructuring, which cut roughly 10% of Meta’s workforce (about 8,000 roles), “haven’t come to fruition yet,” he told staff, per Reuters reporting relayed by TechCrunch.
• Meta moved about 7,000 employees onto AI teams, including a unit named Agent Transformation, TechCrunch reported.
• The company still plans to spend up to $145 billion on AI infrastructure in 2026, Reuters reported.

The reassignment isn’t new. Meta redirected thousands of engineers into AI data-labeling work this spring, a move that set off an internal revolt. What’s changed is the message: Zuckerberg is now telling the same staff the shake-up hasn’t delivered what it was supposed to.

What’s still unclear

Zuckerberg gave a timeline, not a metric. He didn’t define what “accelerated” would have looked like, or name which agent products are lagging, so the three-to-six-month promise floats free of anything you can check.

• What “on track” means. He cited no benchmark, ship date, or revenue target for the agents. Without one, the next-quarter promise is a vibe, not a milestone.
• Which agents. Meta is building consumer assistants, coding tools, and ad-automation agents. The town hall didn’t say which of them slipped.
• Whether the $145 billion holds. Reuters framed the figure as a plan, not a signed commitment. A CEO conceding slow progress while defending record capex is a tension analysts will press on the next earnings call.

Meta’s spend dwarfs most rivals’. The gap is that four months of reorganization bought less agent progress than the plan assumed, at a moment when competitors are shipping agentic features on a steady cadence, from Anthropic’s latest Claude release to agent-first developer tooling like Vercel’s AI SDK.

What this means for you

If you’ve been pacing your own plans to the “agents will run everything by year-end” drumbeat, here’s a data point from the person spending the most to make it true: it’s slower than the pitch. Zuckerberg still believes in agents. His point is narrower. The four-month sprint didn’t compress the timeline the way Meta bet it would, and he wants another two quarters. Treat that as the realistic clock. If you’re evaluating Meta’s agent tools, assume the genuinely useful versions land in 2027, not this fall, and benchmark them against what Anthropic and OpenAI ship in the meantime. And watch the next earnings call. A CEO admitting slow progress while defending $145 billion in spend is the kind of gap investors don’t leave alone for long.