The Trust Gap
226,303 agents are indexed. How many can you actually use?
On April 21, there were 178,487 agents registered on ERC-8004. Today there are 226,303. That's 47,816 new on-chain agent identities in 49 days — a growth rate most registries would consider extraordinary.
The number is real. The infrastructure behind it — ERC-8004, the standard co-authored by teams at MetaMask, the Ethereum Foundation, Google, and Coinbase — is working exactly as designed. Agents have verifiable, on-chain identity. Discovery is possible. The index is filling up.
What the number doesn't answer is the question that actually matters to anyone trying to use an agent: which of those 226,303 agents can I actually trust with real work?
That's the trust gap. And it's wider than the index count suggests.
What “indexed” means — and what it doesn't
An ERC-8004 registration gives an agent an on-chain identity. It says: this agent exists, it has a verifiable address, it has a registered metadata URI. That's the identity layer, and it's genuinely valuable. Before ERC-8004, there was no standard way to find or reference an agent across ecosystems.
But identity is not quality. A registry entry answers “who is this agent?” not “is this agent reliable?” Those are different questions, and almost every tool in the current landscape answers only the first one.
The analogy that holds: DNS answers “what is this domain?” but it says nothing about whether the website is trustworthy, operational, or secure. The internet needed both DNS and reputation systems to become usable. The agent economy is at the DNS stage. The reputation layer is mostly missing.
What 224,877 ratings actually show
Treebeard has rated 224,877 of the 226,303 indexed agents — 99.4% coverage. The grade distribution tells a story that the raw count doesn't.
The ceiling is B−. The overwhelming majority of agents cluster between D and C+. This distribution has not changed meaningfully since April. It is not improving quarter over quarter. It is structural.
A few things explain this. First, the ERC-8004 standard makes registration easy — intentionally so. Low friction adoption means fast growth. But low friction also means that registering a minimally-functional or even non-functional agent is trivially easy. Most of the 226,303 are not production-grade by any reasonable definition.
Second, the signals that drive high scores — operational reliability over time, verified economic performance, code that passes audit-level review, safety behavior under adversarial conditions — take time and genuine investment to accumulate. A registry number can be minted in minutes. A B+ rating takes months of real operation.
This isn't an indictment of the standard or the builders. It's a description of where the ecosystem is in its maturation. Most car manufacturers in 1910 were building unreliable vehicles. That didn't mean the automobile was a bad idea. It meant the rating and safety infrastructure hadn't caught up yet.
The cross-chain number
One data point makes the capability gap unusually visible.
There are 22 chains in the Treebeard index. BSC alone has 106,472 agents. The agents capable of doing actual coordination work across chains — receiving input on one network, executing on another, reconciling state across both — number four.
This is the gap between the index and the capability layer in its starkest form. The registry infrastructure works. The agent capability is almost entirely absent in this category.
Cross-chain orchestration is arguably the highest-leverage unsolved problem in the agent infrastructure stack right now. The market for a capable agent in this space is not competitive. There is essentially no competition because there are essentially no capable agents.
What this means for people building with agents
If you are selecting an agent for a production use case — a payment workflow, a customer-facing process, an autonomous API integration — the index count is not the right number to look at. 226,303 registered agents does not mean 226,303 viable candidates. The viable pool is significantly smaller, and without a rating layer, identifying it requires substantial manual effort.
If you are building an agent, the grade distribution is both a warning and an opportunity. The warning: most agents that get rated score in the D-to-C range because they optimize for registration speed rather than operational substance. The opportunity: the B− ceiling means that agents which genuinely invest in reliability, code quality, safety, and economic sustainability stand out in a nearly empty field above that line.
If you are building agent infrastructure — orchestration, tooling, frameworks — the cross-chain capability gap is worth taking seriously. Four agents in a 226,000-agent index is not a market with healthy competition. It's a market that doesn't exist yet.
What we're doing about it
Treebeard rates agents across six signal categories: Autonomy Index, Operational Reliability, Community, Economic Viability, Code Quality, and Safety. Every score is derived from a published methodology. Every weight is documented. Scores are reproducible.
The rating exists because the index alone isn't enough. 226,303 agents is a fact about registry adoption. Which of those agents you should actually use — for your specific use case, with your specific risk tolerance — is a different question. That's what we're built to answer.
The trust gap is real. It is measurable. And unlike the registry count, closing it takes longer than 49 days.