Meta Buys Manus: Shifting Currents at Meta
What this deal does for Zuck, Wang, and Meta
This is the last post of 2025! Many folks have joined the new exec tier. For the existing exec tier members, don’t forget to check out my 2026 predictions. Follow on LinkedIn
Happy new year!
Meta just announced it’s acquiring Butterfly Effect, the company behind Manus, for an undisclosed amount, rumored to be greater than $2bn (The Information, WSJ).
The company was an early mover in the “AI agent for the average knowledge worker” space, and scaled to about $125M in “ARR” in just 9 months after its viral launch in March.
But despite the impressive topline growth, the company failed to be profitable, had progressively worse customer feedback, and seemed doomed as a “sub-scale” player in the “prosumer AI” market, which has 800-pound gorillas like Claude, ChatGPT, and Gemini as competition.
Therefore, Meta may have just saved Manus with this deal. Obviously, it’s a great outcome for the founders and investors, who hail from both China (ZhenFund, Tencent, and Hongshan) and the U.S. (Benchmark), adding an interesting geopolitical wrinkle (which I’ll get to). But geopolitics aside, this deal is notable for several reasons.
First, Manus is a key customer of Anthropic. In fact, Manus leverages Claude models heavily to drive the “agentic loop”, to the point of being accused as a Claude Wrapper. This complicates things for Anthropic. Suddenly one of its biggest customers (Manus) is becoming the “agent brain” for Meta, which could enter the prosumer market to compete with Anthropic later on. And needless to say, Meta won’t let Claude run Manus’s agent loop forever.
Second, this refuels the debate on whether “agent harness” is suddenly where the value should accrue to, not the model. Essentially, Manus doesn’t train its own foundational models, but crafts environments around them. Zuck clearly thinks that the scaffolding is worth more than models, which fits with the worldview that “LLMs are commodities”. But many disagree with that view, and think agentic harnesses are just temporary workarounds, and not bitter-lesson pilled enough. Which view is right?
But zooming out, the deal has rich competitive implications, and foreshadows what’s to come for Meta’s overall AI strategy.