Mark Zuckerberg isn’t known for admitting things aren’t going well. So when he stood in front of Meta employees at an internal town hall and said AI agent development hadn’t “accelerated in the way” executives had expected, people paid attention.

According to TechCrunch, citing a Reuters report, Zuckerberg also acknowledged that the job cuts Meta made earlier this year weren’t as “clean” as they should have been — and that the anticipated benefits of the company’s new AI-focused structure hadn’t “come to fruition yet.” He said he hoped improvements would arrive in the next three to six months.

That’s a quietly stunning admission from the CEO of one of the world’s most powerful tech companies.

The Cuts That Were Supposed to Usher in an AI Future

Earlier this year, Meta laid off roughly 8,000 employees — about 10% of its corporate workforce — and reassigned another 7,000 people to AI-related roles, including a unit called Agent Transformation, according to TechCrunch. The logic was simple: AI agents were coming fast, humans needed to get out of the way, and Meta needed to reorganize around that reality.

Except the reality didn’t cooperate. The agents didn’t arrive on schedule. And by some accounts, the engineers pushed into Meta’s new AI unit weren’t thrilled about it — TechCrunch notes that several reports have described the months-old unit as a demoralizing place to work.

Zuckerberg reportedly said the cuts happened because leadership “were worried that we weren’t going to move fast enough to adapt.” In other words, the reorganization was driven by fear of falling behind — not by a technology that was actually ready to deliver.

Meta is still betting enormous sums on AI. The company is expected to spend as much as $145 billion on AI infrastructure this year alone, according to TechCrunch.

The Bill Is Coming Due Everywhere

Meta isn’t the only company feeling the strain. Google and Amazon both released sustainability reports this week, and the numbers tell a story the press releases don’t quite spell out.

According to TechCrunch, Google’s total carbon emissions are up 25% compared to last year. Amazon’s are up 16%. Neither company directly blames AI, but both acknowledge that their energy consumption has risen sharply alongside AI use. The bulk of the damage comes from so-called Scope 3 emissions — the pollution generated by things companies buy, like GPUs and data center construction materials — rather than from their own direct energy use.

Building out data centers at AI scale means pouring steel and cement, industries that are among the world’s heaviest polluters. It means buying millions of chips manufactured in Asian factories that still run largely on fossil fuels, using chemicals that are thousands of times more potent as greenhouse gases than CO2. Amazon noted in its report that in 2025 it added more data center capacity globally than any other company, including over 1.2 gigawatts in the fourth quarter alone.

All of that spending — financial and environmental — was predicated on AI delivering transformative results. The returns, so far, are murkier than the pitch decks suggested.

When a Sandwich Shop Mentions AI 22 Times

For a sense of just how inflated expectations have become, consider this: Jersey Mike’s, the submarine sandwich chain, mentioned artificial intelligence 22 times in its recent IPO documents, according to TechCrunch. The company sells subs, not software. Its AI disclosure amounted to little more than “we are beginning to use AI technologies in our business.”

TechCrunch’s Julie Bort noted, drily, that weather was mentioned five times in the same filing. Lightning — an actual documented hazard for a franchise that had a location struck in Texas — wasn’t mentioned once.

The Jersey Mike’s example is funny, but it points to something serious. When investor pressure forces a sandwich company to pad its S-1 with AI references, the hype has become structural. It shapes how companies present themselves, how capital gets allocated, and how executives make decisions — including decisions about which workers to cut and which divisions to build.

What This Moment Actually Means

Zuckerberg’s admission, taken alongside the environmental and financial costs piling up at Google and Amazon, paints a picture the industry has been reluctant to draw itself: AI agents are harder to build than advertised, the infrastructure required is staggeringly expensive and polluting, and the gap between the promises and the products is real.

None of this means AI is a dead end. Zuckerberg himself said he still expects progress. But the era of consequence-free hype may be closing. The bills — in dollars, in emissions, and in credibility — are starting to arrive.