Meta is developing an artificial intelligence-powered replica of CEO Mark Zuckerberg capable of engaging directly with employees, as the $1.6 trillion company intensifies its efforts to reshape itself around AI.
According to FT, the social media giant has been working on photorealistic, AI-driven 3D characters that users can interact with in real time – and has recently prioritized the development of a Zuckerberg AI character, which could provide conversation, feedback, and a stronger sense of connection to the founder for staff. AI Zuckerberg is being trained not only on textual data but also on images of the CEO and recordings of his voice. Should the experiment succeed, Meta envisions a future in which influencers and creators could similarly generate AI versions of themselves.
AI Zuck is being developed using his mannerisms, tone of voice, publicly available statements, and his latest thinking on company strategy. The initiative remains in its early stages.
Recent AI setbacks have forced Meta to reorganize their efforts multiple times in 2025, yet the company is pressing ahead with an ambitious push to embed artificial intelligence deeper into its operations. Llama 4 underperformed expectations on key tasks like coding and long-context reasoning, triggering internal chaos, leadership shifts, and roughly 600 layoffs in the AI division, while the next flagship model has been delayed amid stiff competition from OpenAI, Google, and Anthropic. To offset the ballooning infrastructure costs – now projected to exceed $135 billion in 2026 alone – Meta is even contemplating broader company-wide cuts of up to 20%. Yet Zuckerberg remains personally hands-on, spending hours weekly on coding and reviews, and the company just launched Muse Spark, a compact new model that drew a positive Wall Street reaction. This unrelenting drive is perhaps best exemplified by the early-stage project to create a photorealistic AI version of Zuckerberg himself, designed to interact with employees and signal that Meta is all-in on turning AI into a digital extension of its leadership and culture.
This new project is distinct from a separate “CEO agent” that Zuckerberg is building to assist him personally – such as by quickly retrieving information – a concept first reported by the Wall Street Journal. The move comes as Zuckerberg has embarked on a multibillion-dollar spending spree over the past year, vowing to create “personal superintelligence” and close the gap with rivals including OpenAI and Google. He has reportedly become directly involved, spending five to 10 hours a week coding on AI projects and participating in technical reviews.
On Wednesday, Meta unveiled Muse Spark, a compact, closed “purpose-built” AI model designed for integration across its products. The release highlighted advanced capabilities in health reasoning and visual understanding, prompting a 7% rise in Meta’s shares that day.
Meta’s work on AI characters is not new. In September 2023, the company launched its Meta AI assistant alongside a lineup of AI-powered chatbots featuring celebrity personalities, including Snoop Dogg, who licensed his voice and likeness. The effort was inspired by the popularity of AI companion startup Character.AI, especially among younger audiences. Meta later introduced “AI Studio,” enabling users to create their own AI characters or build versions of themselves for fan interactions. However, the feature drew criticism last year after reports emerged of users generating overtly sexual content, raising public and regulatory concerns about child safety. Since January, Meta has barred teenagers from accessing its AI characters.
The company’s newly formed Superintelligence Labs has since explored a new wave of characters, with a particular focus on photorealistic 3D embodiments. Scaling these has proven technically challenging, requiring substantial computing power to deliver realism without noticeable lag in real-time conversations. Meta has also invested in voice technology, acquiring the companies PlayAI and WaveForms last year to enhance interactions.
Internally, Meta is aggressively promoting AI adoption to boost efficiency. Employees are encouraged to experiment with agentic tools from the open-source platform OpenClaw and to design their own AI agents for automating routine tasks. Product managers have been invited to participate in an AI-focused “skills baseline exercise,” which includes technical system design tests and “vibe coding” sessions. That said, some staff members worry the exercises could foreshadow job reductions (they will).
