Meta thought we’d leave Reality, AI joined us instead.

Its pure irony, instead of humans invading the metaverse, AI agents are invading us and the web

Mark Zuckerberg made a bet on the opposite thing that is actually happening

In October 2021, Mark Zuckerberg stood before the world and declared a new era. ‘Facebook’ was dead. Long live Meta.

The vision was seductive: a billion people inside virtual worlds within a decade, hundreds of billions in digital commerce (hello NFTs!), millions of new creator jobs. “The next platform will be even more immersive,” Zuckerberg promised. “An embodied internet where you’re in the experience, not just looking at it.” We humans would be humans, but in a different dimensionality where we could be whatever and whoever we wanted to be

It was fantastical, but it wasn’t crazy. This is the world as seen through the mind of Ernest Cline, who wrote Ready Player One, and it was adapted into a blockbuster movie by none other than Stephen Speilberg. That story itself was derived from a long line of science fiction, but fully realized in his rendition. For many who were raised in the heyday of Second Life, World of Warcraft, and many MMPORGs; this made a ton of sense. Zuck bought it wholesale, and he had the money to make a bold gamble.

He had been laying the groundwork for years, buying Oculus for $2 billion in 2014, doubling what he paid for Instagram. The metaverse wasn’t a pivot born of desperation. It was the logical extension of everything Facebook had built: connecting people, but more.

And the timing felt right - there HAD to be a sequel to the world wide web as we knew it, it was kind of long in the tooth at that point, since it kicked off hard around 1990. The crypto enthusiasts knew it. The venture crowd sensed it. And Facebook was going to bet on it and build it before anyone could. In a case of confirmation bias, the pandemic forced the physical world of humans online. Remote work became normal. Virtual connections were no longer weird. If there was ever a moment when “teleporting as holograms” to escape physical limitations made sense, this was it.

So Meta went even more all-in. Expanded investments in Reality Labs. Horizon Worlds. Quest headsets… one of us even got one for a birthday present! Billions upon billions in R&D.

The tab? Over $70 billion in losses since 2021, according to Bloomberg. Nearly $5 billion in a single quarter by late 2024.

And then, somewhere in the middle of all that spending, the world changed to what it was always intended to be, in hindsight - enhanced humans bestriding both the physical world and the cyber one.

November 30, 2022: The Day Everything Shifted

ChatGPT launched. Within five days, one million users. Within two months, 100 million: the fastest-growing consumer application in history. Today, 800 million people use it weekly. The explosion caught everyone off guard, including OpenAI. Engineers later admitted they hadn’t expected it to be very successful. But the numbers told a different story. Suddenly, AI wasn’t a research curiosity. It was a product. A tool. Something people wanted, not something they had to be convinced to use.

Google declared “code red.” Microsoft scrambled. And Meta, still bleeding cash on virtual worlds nobody was visiting, faced a choice. The metaverse asked people to leave reality. AI showed up in our reality. One required an expensive headset, a learning curve, social dislocation - an act of faith. The other required nothing but a browser.

The Rise of the Agents - Autonomous AI is finally arriving

The AI story has accelerated fast. Anthropic (founded in 2021 by former OpenAI executives Dario and Daniela Amodei) launched Claude publicly in March 2023. By May 2025, Claude 4 arrived with something different: not just chat, but agentic capabilities. Claude Code and has let people delegate real work (coding, debugging, browser automation) to an AI that showed up in their terminals, their Slack channels, and their actual workflows. OpenAI followed quickly. It also pushed reasoning models. Google released Gemini 2.0 in December 2024 with “agentic” features baked in. The entire industry converged on a new thesis: AI doesn’t need to be smarter. It needs to do things. A fancy thing insiders call “tool calling”.

And the market responded. By 2025, 79% of organizations reported adopting AI agents at some level (Ok ok, let’s maybe call them fancy scripts that fetched things that were hard to fetch, under watchful human eyes.) The global AI agent market hit $7.38 billion, nearly doubling in two years. McKinsey found 23% of enterprises already scaling agentic AI systems, with another 39% experimenting.

The metaverse was supposed to be the future of connection. Agents became the future of life AND work - browsing the web for us, becoming usable software robots that reasoned through new information.

Unscrambling the Metaverse

Meta saw what was happening. In 2023, Zuckerberg declared it the “Year of Efficiency,” code for cutting costs and reallocating resources. The company released Llama, its open-source AI models, positioning itself as a disruptor rather than a follower. Llama reached nearly one million downloads per day by late 2024.

The AI investments accelerated. Capital expenditure guidance for 2025 climbed to $70-72 billion, nearly matching everything Meta had lost on the metaverse. But this time, the money was going to data centers, GPUs, and model training. Meanwhile, Reality Labs kept bleeding. In late 2025, Bloomberg reported Meta was preparing to slash the metaverse budget by 30%, potentially $4-6 billion in cuts. Layoffs followed. The Horizon Worlds team shrank. The Quest hardware unit contracted. In 2026 10% of Reality labs was cut again, drawing down that investment to the lowest level in 5 years.

Meta didn’t and hasn’t abandoned the metaverse entirely. But the pivot has been unmistakable. As one analyst put it: Meta stopped optimizing for quarterly optics and started optimizing for strategic independence. The Ray-Ban Meta Smart Glasses (embedding Llama directly into wearables) became the new bet. Not virtual worlds you escape into. AI that travels with you in the real one.

And let’s not forget Apple bought into this, too. The Apple Vision Pro exists as a testament to how real the taste of the metaverse was, in the mind of Silicon Valley executives. Apple spent over $30 billion on the Vision Pro ecosystem - custom software, custom silicon, custom optics, developer programs, marketing, etc. And it’s estimated that it sold under 1 million units. In comparison, the iPhone sold over 240 million units in 2025 alone.

This Pivot Actually Makes Sense

What Meta and Apple are understanding now is: The metaverse was a platform play. It was a new operating system like Windows, IOS or Facebook, it would need new apps, it would need new app store, it would create lock-in - terms and strategy that had worked well in the past. Build the world, and people will come. But people didn’t come, not at scale, not yet. It was a technology bet looking for a problem to solve (just ask the NFT guys… Oh, too soon?)

The technology was clunky, the value proposition unclear, the adoption curve steep. AI agents are different. They don’t require any great change to customer behavior. They don’t ask you to buy new hardware or learn new interfaces. They meet you where you already are (a chat interface, your email, your code editor, your messaging apps) and they make you faster. Sometimes 100x faster. And this is huge.

The numbers prove it:

• Individuals who get fluent at using Claude or ChatGPT for various workflows, report tremendous productivity increases.

• 66% of companies using AI agents report measurable productivity gains

• Organizations project average ROI of 171% from agentic AI deployments

• 85% of enterprises are expected to implement AI agents by end of 2025

This isn’t hype. It’s adoption at enterprise scale. And Meta, with 4 billion users across its apps, has something no AI-native startup has: distribution. If Meta AI becomes the default assistant for Facebook, Instagram, and WhatsApp, that’s not competing with ChatGPT for premium subscribers. That’s embedding AI into the daily habits of half the planet. “Meta AI is on track to being the most used AI assistant in the world,” Zuckerberg said at Connect 2024. For once, the bravado might be justified. The only real issue to crack for team Meta is monetization. Embedding inside their collection of social apps leaves them at the mercy of the ad market, which AI advances may come along and decimate.

The Inversion Is Complete

So it’s been quite a journey away from the metaverse. In this agentic moment, the inversion is complete. Now it’s the technology learning us, to become more us. More us and faster, because CPUs. No headsets required. The AI doesn’t ask us to come to its reality; it shows up in ours. In our Slack channels, our terminals, our email, doing the work we used to do, in the world we actually live in. When their intelligence and autonomy is paired with autonomous robots, the physical world will be occupied by alien intelligence (that we created, but nevertheless a bit alien), not the reverse.

Ironically, the advent of Transformer-based Artificial Intelligence will accelerate the metaverse because making interesting new virtual worlds will become easier. They are notoriously hard to construct because they have to be handcrafted (detail is important for realism) to be high quality, vs. just procedurally generated. But now we can combine detail and procedural generation as we enter an era of infinite code. The truth is that many of us need the escape of metaverses occasionally. Many MMOs exist and are profitable. But we all need to surface to socialize, work, and play in the real world a lot more than some tech companies imagined.

Meta spent $70 billion trying to get us to leave Earth. Anthropic, OpenAI, and Google are teaching AI to work on Earth. That’s not just a strategy shift; that’s a complete philosophical inversion. And here’s the poetic twist: Meta could keep its rename. “Meta” comes from the Greek for “beyond.” In the metaverse vision, it meant beyond physical reality. In the agentic era, it means beyond human limitations: AI that extends what we can do, where we already are. Same name. Completely different meaning.

But it will also remind everyone of the misfired bet. So the jury is out on whether they will keep that name.

Key takeaways:

1. Meet humans where they are, not where you wish they’d go. Meta asked people to abandon reality. AI agents showed up in their terminals and email. One required faith. The other just worked.

2. AI agents doing real work in the real world is the entire game. The frontier isn’t smarter models—it’s AI that operates autonomously in actual workflows. Embodied intelligence that extends human capability without leaving Earth.

3. Platform plays create lock-in. Tool plays create addiction. The metaverse needed an ecosystem. Agents just need to make you 10x faster. When something works, you don’t need convincing—you need a subscription.

---

🔍 Want to dive deeper?

Check out our book BUILDING ROCKETSHIPS 🚀 and continue this and other conversations in our 💬 ProductMind Slack community and our LinkedIn community.

---

🎧 Prefer to listen? Check our podcast below ↓

🎥 YouTube → Click Here

🎵 Spotify → Click Here

🎙️ Apple Podcasts →Click Here

---

📣 Stay tuned…

After two standout AMAs, we’re excited to welcome Melissa Perri, author of Escaping the Build Trap and CEO of Product Institute, for “Escaping the AI Trap.”
📅 Tuesday, Feb 17 • 9am CST
Details coming soon.

---