Metaverse AI Assistants

The Promise

On October 28, 2021, Mark Zuckerberg announced that Facebook was becoming Meta. The company would bet its future on the “metaverse”—immersive virtual worlds where people would work, play, and socialize through VR headsets. And those virtual worlds would be populated by AI.

The vision was sweeping. AI avatars would serve as guides, companions, and assistants in virtual spaces. They would help you navigate Horizon Worlds, Meta’s social VR platform. They would be NPCs in games, tutors in educational experiences, and colleagues in virtual offices. The metaverse would be alive with artificial intelligences, making the virtual feel vibrant and populated.

Meta’s AI research lab was folded into Reality Labs. The company poured resources into avatar creation, natural language understanding, and embodied AI. The goal: create AI beings that felt native to virtual space, that could see you, hear you, and respond naturally in three dimensions.

The Rise

The spending was unprecedented. Reality Labs burned through $10 billion in 2021, $13.7 billion in 2022, $16 billion in 2023. By 2025, cumulative losses exceeded $70 billion. Meta hired armies of researchers, engineers, and content creators. The company built VR headwear, haptic gloves, and eye-tracking systems.

The AI research was genuinely impressive. Meta developed sophisticated avatar systems, realistic character models, and natural conversation capabilities. Research papers on embodied AI, social robotics, and avatar animation poured out of the company. The technical foundation for AI-populated virtual worlds was being laid.

Zuckerberg personally championed the vision, appearing in VR for announcements and demonstrating avatar interactions that showed AI assistants helping users in virtual spaces. The company rebranded entirely around the metaverse narrative, staking its identity on a future that hadn’t arrived.

The Fall

But the metaverse itself never materialized. Horizon Worlds, Meta’s flagship VR social platform, peaked at a few hundred thousand monthly users—a rounding error for a company with billions of users on its social apps. The virtual worlds were empty. The AI assistants had no one to assist.

The problem was fundamental: VR headsets are uncomfortable, isolating, and inconvenient. People didn’t want to strap devices to their faces to hang out with friends or attend meetings. The compelling use cases—beyond gaming—never emerged. The AI that was supposed to populate the metaverse had nowhere to live.

The pivot came in 2023 and accelerated through 2024-2025. Zuckerberg’s public messaging shifted from “metaverse company” to “AI company.” The Llama large language models became Meta’s flagship AI products—text generators, not avatar assistants. Reality Labs’ budget was slashed by 30%. Layoffs hit the VR division hard in January 2026.

The AI assistants built for the metaverse found new homes as chatbots, as features in Instagram and WhatsApp, as traditional software. The embodied, spatial, immersive AI vision evaporated. $70 billion had been spent preparing AI to inhabit virtual worlds that humans declined to visit.

Warning Signs

• Building for a platform that didn’t exist: Creating AI for VR before VR had mainstream adoption was premature optimization
• Ignoring user signals: Years of weak VR adoption numbers were dismissed in favor of executive vision
• Sunken cost fallacy at scale: Each year of losses justified more investment rather than reassessment
• Technology mismatch: VR hardware limitations made the promised experiences impossible to deliver
• Competitive pressure from the right: While Meta built for VR, AI breakthroughs happened in text and image—the 2D modalities

Epitaph

🪦 $70 billion spent teaching NPCs to wave at empty rooms