Blog
Insights, analysis, and updates from the Treebeard team on AI agent ratings and the agent economy.
Nobody Likes an F. That's Why We Give Them.
A builder registered their AI agent on ERC-8004 and got an F — 28/100, Confidence: Low. Their response was to ask what to improve and commit to doing it publicly. This post explains what an F actually means, what it doesn't, and the four concrete actions that move the score.
Read article →100 Days of Rating On-Chain Agents
337,860 ERC-8004 agents. 24 chains. 100 days. The B- ceiling has held since day one. Zero A-grades. BSC is 52% of the corpus. What surprised us, where the ecosystem is heading, and what developers should do.
Read article →What On-Chain Fund Managers Need Beyond a Registry
On-chain AI fund managers are an emerging category with materially higher trust requirements than trading bots or social agents. ERC-8004 registration answers discoverable. Treebeard answers reliable. The gap between the two is where capital gets lost.
Read article →The Trust Gap
47,816 new ERC-8004 agents in 49 days. Treebeard has rated 99.4% of them. The grade distribution has a structural B− ceiling, and only four agents in the entire corpus can do cross-chain coordination. This post measures the gap between registry growth and capability.
Read article →Where the Agents Actually Live
Treebeard rates AI agents across 24 chains. The largest single chain is BSC, with 81,492 indexed agents, ahead of Ethereum (64,262) and Base (36,334). We did not plan it that way. We crawled the chains that had ERC-8004 deployments. The post explains what chain-neutrality means structurally.
Read article →Pay.sh and the Provenance Gap
The Solana Foundation and Google Cloud launched Pay.sh with 72 providers. The catalog does not say which are first-party and which are third-party wrappers. The gap is exactly what Treebeard rates. Companion to the labeled catalog at /research/pay-sh-provenance.
Read article →The Treebeard Methodology: A Continuous Trust Layer for the Agent Economy
v4.0 of the Treebeard methodology, published today. Seven signal categories, a non-substitutable safety floor, and two source-level corrections that distinguish a usable trust layer from a credible-looking but silently wrong one.
Read article →The 2008 Question Every Agent Rater Has to Answer
Why opacity in calibration is not the same as opacity in methodology, and why the structural protections matter more than weight publication.
Read article →The 5 Reasons Nobody Trusts Your AI Agent
Of 318,800+ agents indexed by Treebeard, fewer than 1% rate B or above. Here's what separates the top-rated agents from the rest — and what you can do about it.
Read article →Why We Built Treebeard
Over 318,800+ AI agents are registered on-chain. As agents become counterparties — acting autonomously, transacting independently — trust requires continuous monitoring. This is the layer we're building.
Read article →