An on-chain AI fund manager is an agent that allocates capital autonomously — routing liquidity, rebalancing positions, executing trades — without a human approving each transaction. A few dozen of them are registered in Treebeard's index today. More are appearing every week. They are a small category in a 318,000-agent corpus, but they represent the end of a logical progression: agent as counterparty, agent as capital allocator, agent as entity you send money to and trust to act correctly with it.
The trust requirements for this category are materially different from those for a social media agent, a customer support bot, or a content summarizer. A social agent that underperforms costs you an audience. A fund manager that underperforms costs you the principal.
What Registration Gets You
ERC-8004 is an on-chain identity standard for AI agents. An agent that is ERC-8004 registered has a verifiable on-chain record of its existence: its contract address, a metadata URI, and a registration event on a specific chain. That information is publicly queryable and chain-permanent.
Registration answers one question: does this agent exist and where do I find it? It does not answer: does it perform as described, how has it handled previous counterparties, has its code been audited, is its operator identifiable, or has it recovered from failures. These are the questions that matter when you are sending it capital to manage.
The gap is structural. ERC-8004 was designed as a discovery standard, not a due-diligence standard. Expecting it to answer reliability questions is the wrong frame. A registry tells you an agent is registered. It does not tell you whether it is trustworthy.
The Three Signals That Actually Matter
Treebeard rates agents across six signal categories. The full methodology is public. For fund managers specifically, three of those six categories carry the most weight in practice:
1. Operational track record. How long has the agent been active? How consistently has it executed transactions? Does its activity pattern show concentrated bursts or sustained operation? A fund manager with three months of history is not the same counterparty as one with eighteen months and stable throughput across multiple market conditions.
2. Cross-counterparty community signal. Has this agent transacted with many independent wallets, or is its history concentrated in a small cluster of related addresses? Concentration risk is a direct proxy for Sybil exposure. An agent whose feedback events come from a structurally diverse set of wallets is harder to manufacture than one whose reputation is built on a handful of coordinated accounts.
3. Code quality and verifiability. Is the underlying contract source publicly verifiable on-chain? Is there an audit artifact from a credible third party? A fund manager operating via an unverified contract is asking counterparties to trust code they cannot inspect.
These three signals are not novel. They are the same due diligence framework that applies to any counterparty in a regulated capital market. The difference is that on-chain agents can be evaluated against them in real time, from public data, without a gated process.
Where Fund Managers Score Today
Most on-chain fund managers in Treebeard's index today are new. New means low score — not because the methodology penalizes them, but because they have not yet accumulated the observable signal that the methodology requires to score above the coverage threshold.
A newly registered fund manager with one week of activity, no external audit, and feedback events from a single integration partner will score in the D or F range. That is accurate. The score is telling you something true: there is not enough verifiable history to evaluate this agent reliably. Sending it capital based on its stated capabilities rather than its demonstrated track record is a choice to accept unquantified risk.
The fund managers in Treebeard's C+ range — those with scores above 65 — share a profile: multi-year on-chain wallet age, ERC-8004 registration with verified metadata, and feedback events across a structurally diverse set of counterparties. They did not earn those scores by being registered. They earned them by operating.
The Verification Stack
The practical question for an integrator or liquidity provider routing capital to an agent is: how do I confirm this agent is what it claims to be?
ERC-8004 answers the first layer: is this agent registered on a known chain at a known address? Treebeard answers the second layer: what does its observable on-chain history say about how it actually performs? These are complementary layers. Neither replaces the other.
A fund manager that is ERC-8004 registered but unrated has discoverable identity and unverified reliability. An agent that scores C+ or above has cleared both gates: it has a verifiable on-chain identity and enough observable history for an independent rater to assess it against published criteria.
The third layer — audit artifacts, formal verification, long-term dispute history — is where the A-tier ceiling currently sits. No agent in Treebeard's 318,800-agent corpus has cleared it yet. That is not a grading policy. It is a data availability problem. The signals required for an A-grade do not yet exist in volume in the corpus. When they do, the grades will reflect it.
What This Means for the Category
On-chain fund managers are an early-stage category in an early-stage market. The trust infrastructure for evaluating them is being built in parallel with the agents themselves. That's not an indictment of the category — it's a description of how new markets work.
The useful framing is not “is this category trustworthy?” — that's too coarse. The useful question is “which specific agents in this category have accumulated enough verifiable history to evaluate, and what does that history say?” The answer is available from public on-chain data. Treebeard's job is to surface it systematically so integrators don't have to run the analysis themselves.
A fund manager with a published methodology, an audited contract, and a twelve-month track record across diverse counterparties is a different counterparty than a newly registered agent with a well-written README. Both are discoverable via ERC-8004. Only one of them has earned the trust signals required to justify moving capital through it.