Meta chief executive Mark Zuckerberg is developing a personal AI agent to help him run the company, according to a Wall Street Journal report published on March 23, 2026. The system is intended to support his daily decisions, surface information and assist with managing an organization whose platforms, including Facebook, Instagram and WhatsApp, reach billions of users.
According to the report, the agent is designed to go beyond retrieving information. It is meant to understand Zuckerberg’s personal context, such as his past activity, priorities and working relationships, so that its suggestions reflect how he actually works rather than generic answers.
A test case for personal superintelligence
The project fits a vision Zuckerberg has described publicly for some time. He has argued that AI delivers its greatest value when it is personalized to an individual, and in 2025 he reorganized Meta’s AI efforts under the banner of personal superintelligence, with the goal of giving every person an AI that helps with their life, work and decisions. He has pointed to 2026 as a critical year for that effort.
Building such a system for himself first lets Zuckerberg act as both the sponsor and the primary test user. Reports describe the agent as something like an AI chief of staff, and note that Meta is simultaneously encouraging employees to use AI agents in their own work, with some of those agents beginning to interact with one another.
Part of a much larger bet
The personal agent is a small piece of Meta’s AI spending. The company has committed tens of billions of dollars to data centers and computing infrastructure, hired aggressively for its Meta Superintelligence Labs unit, and made investments and acquisitions in AI startups to build out its capabilities. Those moves have drawn investor scrutiny over costs, and a working personal agent for the CEO would be a visible demonstration of what the spending is meant to produce.
Echoes of Jarvis
Zuckerberg attempted something similar nearly a decade ago. In 2016 he spent a year building a home assistant he called Jarvis, which could control lights, music and his front door and respond to simple voice and text commands. He described it at the time as a personal challenge, and the project did not go further, in part because the underlying language and reasoning technology was limited.
Current large language models and agent frameworks make a far more capable assistant feasible. Whether the new system becomes a template for products Meta offers to the public is not yet clear, but the report indicates that the company’s CEO is treating a personal AI agent as a practical tool rather than an experiment.
