Why We Built Treebeard

Over 175,000 AI agents are registered on-chain right now. By the time you read this, there will be more.

Some are doing genuinely impressive things — managing portfolios, executing multi-step DeFi strategies, coordinating with other agents without a human in the loop. Others are empty registrations, no activity beyond the developer who launched them. That's fine. We are in the Cambrian explosion phase of the agent economy, and the experimentation is healthy.

The problem isn't the quantity. The problem is what happens next.

Agents Are Becoming Counterparties

For most of software history, programs executed defined instructions. Behavior was predictable. Trust was implicit.

That model is breaking. AI agents now act autonomously, make decisions under uncertainty, interact with other systems, and evolve over time. They handle funds. They execute transactions. They operate without a human on either side of the interaction.

A counterparty is something you rely on to act correctly under uncertainty. When humans are counterparties, you assess their credibility, monitor their performance, and update your trust continuously. AI agents now require the same treatment.

This is not a discovery problem. It is a continuous counterparty risk problem — and the agent economy has almost no infrastructure for it.

The Signal Problem

The trust landscape today is fragmented by design. Agents are deployed across dozens of chains and registries, each with different standards. The signals meant to inform trust decisions are siloed and increasingly vulnerable to manipulation. Reputation systems anchored to a single source are sybil-prone. And when the entity doing the rating has a token, a marketplace cut, or a chain preference, the score is structurally compromised before you ever read it.

We know what happens when rating infrastructure has impure incentives. It ends badly.

The answer isn't a better single source. It's a compositional system — one that ingests signals from multiple registries, multiple chains, and multiple third-party sources, and discounts each for known conflicts before weighting them. We will even rate the raters. Third-party trust signals can be inputs, but only after being evaluated for methodology quality and gaming resistance.

From Lending to Foundation

Here's the backstory. We started focused on lending and borrowing for AI agents. That still seems like a massive opportunity. But we quickly realized: before any financial primitive can work between agents, foundational trust infrastructure has to exist first. You can't lend to an agent you can't evaluate continuously. You can't build a borrowing protocol without a credit signal that updates.

Treebeard is that layer. Not a static report card — continuous trust infrastructure for developers.

What Treebeard Actually Does

We index agents across multiple EVM chains, aggregate signals from on-chain registries and data partners, and produce composite trust assessments across six signal categories: Economic Viability, Operational Reliability, Autonomy Index, Code Quality, Safety, and Community & Ecosystem.

Every rating is explainable. Methodology published in full. No sponsored listings. No token. No payment from rated entities.

Ratings move as new signals arrive. That's the point. Trust is not static, and a trust system that only updates quarterly is not a trust system — it's a snapshot.

Today, fewer than 7% of indexed agents score C- or above — and fewer than 1% reach C or above. That reflects the reality of a young ecosystem in its experimentation phase. Publishing that honestly is more useful than pretending otherwise.

Built for Developers

Developers can spend months building agents. If those agents don't ultimately deliver or attract users, it's expensive in an already competitive space.

Treebeard's ratings don't just tell you where your agent stands — they tell you why it scored the way it did and what you can do about it. Less report card, more diagnostic. Over time, Treebeard should help developers improve their product, find their audience, and build agents people actually trust.

The Infrastructure Is Arriving

This matters more than it might seem. ERC-8004 gives agents universal, permissionless identity. ERC-8183 enables verifiable agent commerce. And x402 — ratified this week as a Linux Foundation standard, with founding members including AWS, Google, Microsoft, Stripe, and Visa — is the payment rail agentic commerce will run on.

These are no longer experimental protocols. They are infrastructure. The agent economy is not a future state. It is being built now, by serious institutions, at scale.

The trust layer that sits above that infrastructure is what Treebeard is building. Every enterprise that plugs into x402 will eventually face the same question: which agents do I allow to transact on my behalf? That is the question Treebeard is designed to answer.

Collaborate With Us

We can't build this alone.

Developers: Submit your agent. Get actionable feedback. A better-rated agent earns more trust.

Protocol teams and marketplaces: Integrate Treebeard ratings into your flows. The API is free for read access.

Data partners: If you run a security scanner, verification service, or agent registry — your data could strengthen our ratings. We're actively seeking signal sources across identity, security, behavior, and payment layers.

Researchers: Rating autonomous agents is an open methodology problem. We publish our approach and welcome challenges.

The agent economy needs neutral referees. We're building one. If any of this resonates, reach out: patrick@treebeardai.com