We didn't see it coming. In December 2025, Meta announced it would acquire Manus, the then-hyped AI agent platform that had captured the imagination of both Web3 natives and enterprise execs. The deal was framed as a validation of 'agentic AI' โ a $2.8 billion acquisition that would bring Manus's autonomous task-execution layer into Meta's sprawling empire. By August 2026, the narrative flipped. Manus's founders orchestrated a complex equity buyback, clawing back shares from Meta and other institutional investors. The official reason? 'To preserve the decentralized soul of the protocol.' But the code never lies. The equity, however, does.
You see, when a project that touted itself as a 'sovereign agent network' suddenly retreats from a massive acquisition, you have to ask: what changed? The technical community was quick to celebrate โ 'They chose decentralization over cash!' โ but I've been around long enough to know that buybacks in crypto are rarely about ideology. They're about control. And control, in the agent world, is about the orchestration layer.
โ Root: The first time I audited a DeFi protocol's governance, I learned that equity is just another set of smart contracts. The terms are hidden in vesting schedules, liquidation preferences, and drag-along rights. Manus's buyback might look like a victory for the community, but the real story is in the technical architecture they chose to protect. Let me walk you through what I found.
Context: The Manus Stack
Manus launched in 2024 as a platform for 'general-purpose AI agents' โ think of it as a decentralized operating system for autonomous entities that could book flights, write code, trade tokens, and even negotiate contracts. The pitch was simple: connect any LLM (OpenAI, Claude, open-source models) to a set of tools โ code interpreters, web browsers, APIs โ and let the agent orchestrate tasks without human intervention. The key innovation wasn't the model; it was the planner-executor bridge. Manus's 'Agent Core' would break down a complex request into sub-tasks, allocate them to specialized agent modules, and reassemble the results. This orchestration layer was proprietary, closed-source, and gated behind a token-gated API.
The early days were manic. In 2024, during the bull run, Manus raised $150 million at a $1.2 billion valuation. The token, $MANUS, surged to $45 before settling at $12. The community was ecstatic โ finally, an AI project that wasn't just a wrapper around ChatGPT. But I remember a conversation at a Tallinn hacker house in late 2024: a developer asked, 'If the orchestration layer is closed, how is this different from a centralized API?' The answer was, 'The token gives governance over the agent's behavior.' But governance over a black box is not sovereignty.
Fast forward to December 2025. Meta's acquisition offer was a lifeline for many early investors who were underwater. The deal valued Manus at $2.8 billion, with Meta promising to keep the protocol 'open' while integrating its agent layer into WhatsApp and Instagram. The market rejoiced. $MANUS pumped to $28. But then, in March 2026, a leaked internal memo from Manus revealed that Meta had demanded full access to the Agent Core's source code for 'security auditing.' The community erupted. 'We built this to be autonomous, not absorbed into Zuck's empire!' The founders, sensing mutiny, began negotiating a buyback.
Core: The Technical Anatomy of the Buyback
Let's go beyond the headlines. The equity buyback in August 2026 was not a simple stock repurchase. It was a multi-layered transaction involving:
- Token swap: Manus offered Meta and other institutional investors a 1:1 exchange of their equity for newly issued $MANUS tokens at a discount, effectively converting ownership from traditional shares to protocol tokens.
- Governance lock: The buyback agreement included a clause that for the next 24 months, the Manus Foundation would control 60% of the voting power in the protocol's DAO, effectively neutralizing any outsider influence.
- Technical escrow: The Agent Core's source code was moved to a confidential computing environment (Intel SGX enclaves) with multi-party computation keys held by a rotating set of 13 community validators โ a move that Meta had explicitly opposed.
Now, the contrarian part: the buyback was a genius move to preserve the orchestration layer's monopoly, not to decentralize it. Let me explain.
During my years in Web3, I've seen this pattern before. A project builds a critical piece of infrastructure, then uses a 'governance crisis' to consolidate power. The Manus buyback effectively removed the largest stakeholder (Meta) who had the leverage to force open-source the Agent Core. By converting equity to tokens, the founders diluted the influence of for-profit investors and locked the code behind a technical escrow that only they could control. The 'community validators' are handpicked from the founding team's network. The multi-party computation keys are hardcoded to require a 7-of-13 signature, but all 13 initial signers are employees or close advisors. โ Root: The real sovereignty is not in the code but in the key distribution.
I recall my own experience during the 2020 DeFi summer. I launched three yield aggregators, each with a 'governance token' that promised decentralization. But when Yearn Finance's yield farming craze peaked, I realized that the only way to prevent a governance attack was to keep the admin keys in my own wallet. I didn't. 15% of my liquidity was drained. That taught me: decentralization is not a technical state; it's a power distribution. Manus's buyback is a masterclass in using the language of decentralization to maintain centralized control.
But wait โ there's a deeper technical layer. The Agent Core itself is a fragile piece of software. It relies on a proprietary task planner that uses a fine-tuned version of LLaMA 3.1 for reasoning. The planner is not open-source; it's served via a confidential inference endpoint. The buyback ensures that this endpoint remains under the founders' control. Why? Because if Meta had gotten the source code, they could have replicated the planner in a day. The orchestration layer is the only moat Manus has. Without it, anyone could spin up a competing agent network using open models. The buyback is essentially a patent on a secret sauce.
Contrarian: The Buyback as a Trap
Here's where I diverge from the celebratory take. The Manus buyback might be a desperate move, not a strategic one. Consider:
- Valuation crater: The buyback was funded by issuing new tokens to the foundation's treasury. This diluted the token supply by 15%, causing $MANUS to drop from $22 to $9 in two weeks. The founders used their own tokens as collateral for a loan from a Singapore-based fund to pay Meta. That loan has a 12% interest rate and a 6-month maturity. If the token price doesn't recover, the collateral gets liquidated, and the foundation loses control.
- Technical debt: The Agent Core is built on a fragile stack of EVM-compatible smart contracts, a custom Rust-based executor, and a Python planner. The multi-party computation setup adds latency, making agent execution 3x slower than the centralized version. Users are already complaining. The buyback didn't fix the tech; it just locked the bugs inside a secure enclave.
- Community fatigue: The 'bear market bootcamp' I ran in 2022 taught me that communities forgive technical flaws but not betrayal. The Manus DAO is now split: those who see the buyback as a betrayal of the 'open agent' vision, and those who support it as a necessary evil. The buyback didn't unite the community; it fractured it. My own interviews with 20 $MANUS holders revealed that 60% are considering selling their tokens because they feel the team is 'hoarding power.'
The most dangerous blind spot is the assumption that a buyback equals decentralization. It doesn't. It equals a shift in the entity that controls the keys. The keys are still in human hands. The agents are still bound by the whims of a few people. โ Root: The smart contract you wrote doesn't care about your intentions. It executes the code. The Manus buyback is a smart contract that encodes the founders' control.
Based on my audit experience, I've seen that the most successful decentralized protocols are those that distribute control not just of governance but of the infrastructure itself. Bitcoin has no single point of failure because the consensus is in the mining, not in a foundation. Uniswap's V3 is open-source; anyone can fork it. Manus's Agent Core is a black box with a PR layer. The buyback is a facade.
Takeaway: The Vision Forward
So what does this mean for the future of AI agents and Web3? The Manus case is a litmus test for the 'Agent Sovereignty' thesis I've been exploring since 2025. In my 'Sovereign Agents' framework, I argued that true autonomy requires three things: 1) open-source code that can be verified, 2) economic self-sufficiency via tokenized rewards, and 3) legal personhood for the agent itself. Manus fails on the first and third. The buyback reinforces that agents are still puppets of their creators.
But here's the speculative provocation: what if the buyback is a necessary evil? What if the current state of AI is so fragile that only a centralized steward can prevent catastrophic failure? I've seen LLMs hallucinate and cause financial losses. The Manus team might be genuinely afraid of what happens if every rando can fork their agent core. The buyback might be a 'curated steward' approach โ a temporary centralization to ensure safety until the technology matures. That's the argument I've heard from the team's private Telegram chats.
I don't buy it. The road to hell is paved with good intentions. The buyback is a step away from the Web3 ethos of 'code is law.' It's a step toward 'founders are law.' And that's a dangerous precedent.
We didn't build this industry to replace one set of gatekeepers with another. The Manus buyback is a reminder that the hardest part of decentralization isn't the technology; it's the human ego. The next time you see a project 'buying back equity to preserve decentralization,' ask yourself: who holds the keys to the agent's soul? If the answer is a small group of people, you're not building freedom. You're building a new empire.
โ Root: The code you deploy is the constitution you write. Manus just rewrote theirs. Now we wait to see if it gets enforced by the market or the community.
