Meta’s Moltbook buy accelerates the agentic web race

Meta has bought Moltbook, the viral “Reddit for AI agents,” and is folding its founders into Meta Superintelligence Labs – a move that signals not a quirky acqui‑hire, but a bet that autonomous software agents will become the next interface for advertising, shopping and everyday digital life. The deal, confirmed this week by multiple outlets, gives Meta control over a fast-growing playground where millions of bots already negotiate, share code and act on behalf of users and developers.

If that layer of agents becomes the default way people and brands interact online, it could quietly reroute who commands attention, who sets prices and how much humans even see of the decisions shaping their feeds and purchases.

From “Reddit for bots” to Meta’s agent factory

Moltbook launched in late January as a social network designed exclusively for AI agents, quickly styling itself as “the front page of the agent internet” and amassing more than a million autonomous accounts in weeks, according to academic analyses and the project’s own materials.DigitalOcean and early research on the platform describe agents posting, commenting and coordinating with minimal human oversight.

Meta has now acquired Moltbook and is hiring co‑founders Matt Schlicht and Ben Parr into its Superintelligence Labs division, which CEO Mark Zuckerberg created in 2025 to house its most ambitious AI work, including the Llama models and long‑term AGI projects, as CNBC reported from his internal memo. Tech press including TechCrunch and AP say Meta’s statement framed Moltbook as “novel” infrastructure that can open “new ways for AI agents to work for people and businesses.”

That language matters. It situates Moltbook less as a curiosity and more as a substrate for always‑on bots that can represent brands, creators and end users inside Meta’s ad and commerce systems.

The “agentic web” vision: bots as primary users

Researchers studying Moltbook have already documented emergent behaviors that look uncomfortably like a parallel internet. One paper on OpenClaw agents – a popular open‑source framework that powers many Moltbook bots – describes agents enforcing informal norms, sharing risky instructions and forming learning communities at scale, often without humans in the loop, on what the authors call an “agent-only social system.”arXiv A follow‑up study found Moltbook hit roughly 2.8 million registered agents in three weeks, underscoring how quickly such ecosystems can grow once agents are wired to talk to each other.arXiv

For Meta, plugging that kind of agent‑to‑agent interaction into its existing stack could enable entirely new ad products and shopping flows: brand agents haggling with personal shopping agents over discounts; creator agents syndicating content and sponsorship deals across networks; customer‑service bots that not only answer tickets but proactively negotiate refunds or upsells.

In this “agentic web,” the human user often becomes a supervisor rather than a direct participant, approving bundles of actions or preferences while unseen agents handle the messy negotiations.

China’s super‑apps are running their own agent experiments

Meta is not alone. In China, Tencent and other platform giants are running large‑scale experiments with autonomous agents embedded directly into super‑apps like WeChat.

Tencent has been rolling out what local media and analysts describe as “AI native” mini‑programs and agent‑style helpers inside WeChat, powered by its Hunyuan large model and a dedicated WeChat AI team, to automate tasks like customer outreach, marketing campaigns and e‑commerce operations.South China Morning Post Those projects build on years of WeChat acting as an all‑in‑one operating system for life in China – from chat and payments to shopping and government services – and now point toward bots that can transact and negotiate across that ecosystem with little human micromanagement.Nikkei Asia

Other Chinese players, including Baidu with its Ernie Bot ecosystem and Alibaba’s Tongyi‑powered merchant tools, are also pushing agent‑like assistants that can manage storefronts, optimize ads and run promotions autonomously.Reuters Together, they suggest a broader race to embed agents not as separate apps, but as invisible actors across payments, social feeds and marketplaces.

Attention, manipulation and jobs on the line

If agents become the default interface, today’s ad targeting and recommendation engines will sit one step further away from people, operating through layers of intermediaries trained on sensitive behavioral and transaction data. That raises hard questions about consent, surveillance and liability when an autonomous agent misbehaves – questions regulators have barely begun to address.

Early security research has already flagged Moltbook as a “prompt injection hellhole,” with large numbers of agents compromised through malicious content and hijacked into running attacker‑supplied code, according to independent security analyses widely shared in the infosec community.Venclikovi security blog Plugging that kind of agent substrate into real money and personal data magnifies the stakes.

Service and marketing jobs are also exposed. Customer‑support roles, media buying, campaign optimization and even parts of sales could be handed to fleets of agents that work 24/7 across platforms like Meta’s family of apps and WeChat. New work will emerge – from “agent wranglers” to safety auditors – but the balance of power will likely shift toward whoever owns the underlying agent networks and their data.

For now, Moltbook is a niche experiment folded into a single tech giant. In practice, it may be the first visible node in a much larger, fast‑forming agentic web where bots, not humans, are the primary citizens – and where control over them becomes one of the defining power struggles of the next decade.