Mark Zuckerberg is building an AI agent designed to assist him with his duties as chief executive of Meta.
According to reporting by the Wall Street Journal, the system remains in development but already functions as an on-demand information tool that allows the chief executive to access data faster than traditional hierarchical channels would permit.
The significance of this move extends beyond a single executive’s convenience. It represents a striking admission about how large technology organisations actually operate: that valuable information often gets lost or delayed somewhere between individual teams and the C-suite.
By automating this retrieval process, Meta is acknowledging that the friction of information flow, the layers of communication, the need to coordinate across departments, presents a genuine efficiency problem worth solving with artificial intelligence.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
Zuckerberg has been unusually transparent about his ambitions for AI to reshape how Meta functions.
During the company’s fourth-quarter earnings call on 28 January, he declared that this year would be when “AI starts to dramatically change the way” Meta works.
More provocatively, he suggested that “projects that used to require big teams now be accomplished by a single, very talented person.” That vision appears to be moving from aspiration into practice.
Meta employees already have access to internal AI tools that foreshadow the broader reorganisation underway.
MyClaw grants workers access to internal files and chat logs, enabling them to communicate with colleagues or AI agent counterparts without navigating bureaucratic handoff points.
A second system, known as Second Brain, builds on Anthropic’s Claude infrastructure and functions as a personal chief of staff, organising tasks, surfacing insights, and streamlining the retrieval of institutional knowledge.
These tools are delivering measurable results, at least by Meta’s own accounting.
Susan Li, the company’s chief financial officer, reported during the same earnings call that output per engineer has risen 30 percent since the start of 2025, primarily driven by AI coding agents.
“Power users”, staff who have fully embraced the new AI systems, have seen output increases of 80 percent year on year. Such gains suggest that AI integration is not merely symbolic but is genuinely accelerating how much work individual employees can complete.
The financial implications are substantial.
Meta has forecast capital expenditure of $115 billion to $135 billion for 2026, nearly double the $72 billion it spent in 2025.
This dramatic escalation reflects a company betting heavily that AI infrastructure and tools will generate returns sufficient to justify the unprecedented investment.
To that end, Meta acquired Manus, a general-purpose AI agent developer, for $2 billion in December 2025. The company also established Meta Compute, a new top-level organisation led by Santosh Janardhan and Daniel Gross, the latter brought in from Safe Superintelligence, signalling both the priority and the resources being devoted to this direction.
Building an AI agent capable of assisting a chief executive is a different challenge from deploying coding assistants or general information tools.
A system designed to help run a company must navigate competing priorities, weigh strategic decisions with incomplete information, and develop contextual understanding of organisational politics and human relationships.
These tasks strain even sophisticated AI systems.
Zuckerberg’s approach appears pragmatic: rather than attempting to create an AI agent that independently makes executive decisions, he is building one that accelerates his access to information and accelerates his ability to process it.
The agent retrieves answers that would ordinarily require coordination across multiple teams and layers of employees, the sort of logistical overhead that consumes executive time without producing strategic value.
By automating this layer, Zuckerberg is claiming back time for the decisions that machines cannot yet reasonably make.
Whether this experiment succeeds, if an AI agent can genuinely function as an executive assistant, will likely influence how other technology leaders approach their own organisations.
For a decade, the technology industry has discussed the potential for AI to flatten hierarchies and reduce managerial overhead. Meta is now testing whether these claims hold water. The stakes are substantial.
If Zuckerberg can demonstrably accomplish more with the same time investment by relying on an AI agent, the incentive for other large organisations to pursue similar strategies will be overwhelming.
If the experiment stalls or produces diminishing returns, it may signal that the flattening-of-organisations vision remains further away than current rhetoric suggests.
For now, Meta is investing tens of billions of dollars on the bet that AI will transform how it operates. The question is whether having an AI agent in the executive suite is a temporary productivity hack or the first visible sign of a more fundamental reorganisation to come.
