Mark Zuckerberg is building an AI agent that functions as a personal executive assistant, retrieving information that would normally require going through multiple layers of people and teams at Meta. The Wall Street Journal reported on March 22 that the tool is already operational, though still under development.
The agent runs on Anthropic's Claude infrastructure. It sits alongside two internal Meta tools already in use by employees: MyClaw, which provides access to work files and colleague communications, and Second Brain, described internally as an "AI chief of staff" for accelerating project work.
This is not a side experiment. On Meta's January earnings call, Zuckerberg laid out the strategy directly: "We're investing in AI-native tooling so individuals at Meta can get more done, we're elevating individual contributors, and flattening teams." The goal is to make Meta's 78,000-person organization operate with the efficiency of a much smaller, AI-native startup.
The practical implication of a CEO agent that retrieves answers directly is that it compresses the information chain. Instead of a question traveling down through VPs, directors, and managers before an answer travels back up, the agent pulls from internal systems and delivers it. That is useful for a CEO. It is less reassuring if you are one of the people in that chain.
Reuters has reported that Meta may cut up to 20% of its workforce, though Meta has called that figure "speculative." But Zuckerberg's own framing - flattening teams, elevating individual contributors - describes exactly the kind of organizational compression that reduces headcount in middle management.
The fact that Zuckerberg built this on Claude rather than Meta's own Llama models is a notable detail. Meta has invested billions in Llama and positioned it as competitive with the best closed models. When the CEO of that company reaches for a competitor's product for his own daily work, it says something about where he thinks the capability gap still sits.