Zuckerberg Creates "CEO-Exclusive AI Agent," Meta's 78,000-Person Company Begins Letting AI Agents "Socialize" With Each Other

Internal Meta is quietly undergoing an organizational revolution: employees’ AI agents are beginning to communicate with each other, engineers’ productivity has increased by 30%, and Mark Zuckerberg himself is creating a dedicated “CEO Agent” to bypass bureaucratic layers and get answers directly.
(Background: No models, criticized as mere shells, how Manus is breaking AI industry barriers with “incremental thinking” and joining Meta)
(Additional context: Meta’s shocking acquisition of “AI-specific community” Moltbook! The founders of both companies join the Meta Superintelligence Labs)

Table of Contents

Toggle

• Zuckerberg’s CEO Agent: Skip bureaucracy, get answers directly
• My Claw, Second Brain: Every employee has their own agent
• AI usage included in performance evaluations, engineers’ output surges 80% in a year
• Acquisitions of Manus and Moltbook: External landscape of the AI agent ecosystem
• Organizational flattening is truly happening

On Meta’s internal message board, there’s a group called “agent-to-agent”—as the name suggests, a place where employees’ AI agents communicate with each other. This isn’t science fiction; it was revealed in a Wall Street Journal report on March 23. A tech giant with 78,000 employees is redefining the very meaning of “work” at a faster pace than outsiders imagine.

Zuckerberg’s CEO Agent: Skip bureaucracy, get answers directly

According to WSJ, Zuckerberg is developing a “CEO Agent” to handle his daily tasks. Still in development, its main function is to help him “access information faster”—specifically, to query internal systems and retrieve answers that previously required multiple layers of communication.

This detail is worth pondering. The CEO should have the fastest access to the information they need, but in a company of nearly 80,000 people, even the CEO often has to go through organizational processes to get specific data or decision context. The AI agent essentially replaces these middle steps.

Zuckerberg’s goal isn’t just for his own use. He hopes that eventually everyone inside and outside the company will have their own personal AI agent.

My Claw, Second Brain: Every employee has their own agent

Currently, Meta employees can use a personal agent tool called My Claw. My Claw can access employees’ chat logs and work files, and “represent” them in communication with colleagues—or more precisely, communicate with colleagues’ AI agents.

Here emerges a concept never before widely implemented in corporate settings: agent-to-agent interaction. When your agent is negotiating with my agent, the human user steps back, and the flow of information becomes much faster and more efficient.

Another tool, Second Brain, is an internal file indexing and query system that allows employees to quickly locate scattered internal knowledge. Together, these form the infrastructure Meta is building for “AI-augmented employees.”

AI usage included in performance evaluations, engineers’ output surges 80% in a year

More indicative is that Meta has incorporated AI tool usage into employee performance assessments. This isn’t just encouragement; it’s a systemic push—using AI effectively directly impacts your performance review.

Numbers speak: Meta CFO Susan Li states that since early 2025, overall engineer productivity has increased by 30%, mainly driven by AI programming agents; among heavy users, the increase reaches 80%.

This echoes Zuckerberg’s previous statement: “Projects that used to require large teams can now be done by a single talented person.” This is no longer just a vision but a reality Meta is realizing with data.

Acquisitions of Manus and Moltbook: External landscape of the AI agent ecosystem

Meta’s AI agent ambitions extend beyond internal tools. In December last year, Meta acquired Chinese AI startup Manus for about $2 billion. Manus is known for autonomously completing complex multi-step tasks. Earlier, Meta also acquired the community platform Moltbook, focused on AI agent interactions. Both founders have joined Meta’s Superintelligence Labs.

From internal tools (My Claw, Second Brain, CEO Agent) to external acquisitions (Manus, Moltbook), Meta is building both the supply and demand sides of the AI agent ecosystem—creating tools to run agents and exploring how agents form ecosystems.

Organizational flattening is truly happening

Putting it all together, a deeper structural shift emerges beyond “AI productivity tools”: Meta is replacing organizational layers with AI agents.

In the past, middle management existed partly to relay information, coordinate, and synthesize. When the CEO’s agent can directly access the answers they want, and employees’ agents can negotiate directly with each other, organizational flattening is no longer just a management philosophy but a technically achievable reality.

For the crypto and Web3 industries, this trend is equally noteworthy. Decentralized protocols have long promoted “disintermediation,” and Meta is doing the same internally with AI agents—just via a different path. As the “agent economy” moves from concept to daily operations in the world’s largest tech company, the normalization of this shift has only just begun.