Meta's Newest Product: An AI Zuckerberg That Won't Stand You Up

Remember that horrifying Horizon Worlds selfie from August 2022? The legless, dead-eyed cartoon standing before a tiny Eiffel Tower that the internet absolutely eviscerated? Even Meta's own employees reportedly refused to use it. Well, Mark's decided to skip the middleman entirely. Nothing says "we're taking this seriously" like replacing your metaverse avatar with a digital twin who actually has legs—or at least the illusion of them.

According to a Financial Times report, Meta is now building a photorealistic, AI-powered 3D version of its CEO designed to hold real conversations with employees on his behalf. Four people familiar with the matter told FT that Zuckerberg is personally training and testing the system, feeding it his mannerisms, vocal patterns, public statements, and recent thoughts on company strategy. The goal? Make employees "feel more connected to the founder" through an AI that talks like him, thinks like him, and never has to cancel a one-on-one meeting. Nothing builds company culture like a chatbot that can't ghost you on Slack.

The project is being led by Meta's newly formed Superintelligence Labs. Scaling the tech has proven difficult—it requires enormous computing power to keep interactions realistic and lag-free. Meta last year acquired two voice companies, PlayAI and WaveForms, as part of that push. The company's projected capital expenditure for 2026 sits between $115 billion and $135 billion, nearly double last year's figure. That's enough compute to simulate consciousness—or at least simulate Mark caring about your quarterly OKRs.

Last week, Meta released Muse Spark, the first model from its Superintelligence Labs—a compact, purpose-built system with capabilities in health reasoning and visual understanding. Shares jumped 7% on the announcement. The market clearly believes in the vision, or at least believes in the capex spending spree. Bullish behavior from investors who probably remember the billions flushed down the Reality Labs toilet.

Inside the company, employees are being pushed to embrace AI tools and build their own agents using open-source software called OpenClaw. Product managers have been handed a "skills baseline exercise" that includes system design tests and, yes, "vibe coding." Nothing says "future of work" like being evaluated on your ability to code by vibes alone. Elon Musk would be proud—or laughing, probably laughing.

The contrast with the metaverse era is stark. As previously reported, Horizon Worlds was in a self-declared "quality lockdown" while its own team was barely logging in. Reality Labs burned through billions every quarter—$10.2 billion in 2021 alone—before Zuckerberg quietly pivoted. That dead-eyed cartoon became the defining image of that failure. Peak comedy: the virtual world nobody wanted to visit was being managed by people who wouldn't even visit it themselves.

Now the bet is on something that looks and sounds like the real thing—to either make employees feel more connected to leadership, or just more supervised by it. The difference between a digital twin and a surveillance tool is, apparently, a $115 billion budget and some tasteful voice synthesis. At least this time the Zuckerberg avatar has legs—and a pension plan.