Meta AI agents are not moving as fast as Mark Zuckerberg expected, and that admission is now raising bigger questions about how quickly Big Tech can replace human work with automated systems. According to Reuters, Zuckerberg told employees during an internal town hall on Thursday that the development of AI agents over the past four months had not “accelerated in the way” executives expected.
That is a major statement coming from the same company that has aggressively reorganized itself around artificial intelligence. Zuckerberg also told staff that Meta’s restructuring was not as “clean” as it could have been and that the company’s bets on the new setup have not “come to fruition yet.” In plain terms, Meta moved fast, cut deep, and pushed thousands of people into AI-related work, but the payoff has not arrived on schedule.
The moment lands awkwardly because Meta has been operating like AI agents were right around the corner. These systems are supposed to complete tasks on behalf of users, assist employees, automate workflows, and eventually make companies leaner. The pitch has been that AI will not just answer questions, but take action. The problem is that getting software to reliably do real workplace tasks is proving much messier than the boardroom hype made it sound.
Meta laid off about 10 percent of its global workforce and reassigned roughly 7,000 employees to AI-focused teams in May. Many employees were told they had been selected for a new AI initiative spun up directly by Zuckerberg, with some workers moving into Applied AI and others joining teams focused on AI agents, including Agent Transformation Accelerator and Agent Data and Optimization.
Those moves are part of why Meta AI agents are under such a bright spotlight right now. Some employees described the reassignment process as being “drafted,” while others expressed confusion over what the new roles would actually involve. Meta also told some employees that transfers into new AI work were not optional, which added to concerns that the company’s once flexible culture was being replaced by a more top-down approach.
Zuckerberg explains that executives feared Meta would not move fast enough to adapt to the changing tech landscape. That fear is not coming from nowhere. OpenAI, Google, Anthropic and other major AI players have been battling for dominance, and Meta has spent the last few years trying to prove it can lead instead of trail. According to Meta, the company rolled out Meta AI across Facebook, Instagram, WhatsApp and Messenger with Llama 3 in 2024, positioning it as a free assistant built directly into apps people already use.
Meta’s AI ambitions go even deeper than chatbots. According to Meta, Zuckerberg argued in 2024 that open-source AI was the path forward and announced Llama 3.1 405B as what Meta called its first frontier-level open-source AI model. That strategy has always been about scale. If Meta can make its AI models widely used, and then plug those models into its social platforms, ads business, workplace tools, and smart glasses, the company can build an AI ecosystem that touches billions of users.
But the money behind that vision is enormous. The company expects 2026 capital expenditures, including finance lease payments, to land between $125 billion and $145 billion. That raised spending forecast rattled investors, especially as Meta also confirmed May layoffs and continued pouring billions into AI infrastructure.
That is the part that makes this bigger than a tech update. Meta AI agents are being presented as the future of work, but workers are already feeling the consequences before the technology has fully delivered. Meta’s AI-centered restructuring sparked internal criticism after the company laid off 10 percent of its workforce, transferred nearly as many employees to new units, and introduced mouse tracking software that some workers viewed as training their own replacements.
The mouse tracking issue has become another pressure point. Meta CTO Andrew Bosworth told employees at the same town hall that a review found no employee data had been included in AI training after a recent data security incident tied to the software. Bosworth also said that if the program returns, it will be opt-in, after employees were previously told there was no way to opt out.
So why is this happening? The simple answer is that AI agents are much harder to build than AI demos make them look. According to an arXiv paper titled AI Assisted Fixes to Code Review Comments at Scale, researchers working with Meta-related code review systems found that one safety trial initially made reviewers more than 5 percent slower before the design was changed. According to another arXiv paper on experienced open source developers, AI tools in a 2025 randomized trial increased completion time by 19 percent for the developers studied, even though the developers expected the tools to make them faster.
That does not mean AI is useless. It means the workplace version of AI is not magic. Software engineering involves much more than writing code, including maintaining systems, understanding context, debugging and preserving security. That same reality applies to other corporate tasks. Agents need access, memory, judgment, safeguards, clean data and an understanding of company-specific work. Without that, they can create more review work instead of less.
Zuckerberg is still not backing away from the bet. He told employees he expects Meta to see more significant benefits from its AI investments within the next three to six months. The billionaire previously described seeing small teams use AI to build products that would have once required dozens of people and months of work, while Meta CFO Susan Li said the company does not know what the optimal size of a future AI-powered company will be.
That uncertainty is exactly why this story matters. Meta is not a struggling startup trying to cut corners. It is one of the most powerful companies in the world, with billions of daily users and the money to build huge AI infrastructure. If Meta AI agents are still moving slower than expected after layoffs, reassignments, internal pressure, and a spending plan that could reach $145 billion this year, then the rest of corporate America should be careful about treating AI replacement as a done deal.
For employees, the message is even sharper. Meta’s latest town hall suggests the future of work is not arriving cleanly or evenly. Companies may be ready to restructure around AI before AI is ready to carry the workload. Zuckerberg’s admission does not end Meta’s AI race, but it does reveal the gap between the promise being sold and the reality employees are living through right now.
