The 5 Reasons Nobody Trusts Your AI Agent

You built an agent. You registered it on-chain. Maybe you even shipped real functionality — portfolio management, cross-chain settlement, developer tooling. The code works.

But nobody's integrating it. Nobody's recommending it. Nobody trusts it yet.

Today we index over 175,000 agents across 14 chains and rate them using a six-category signal methodology. After analyzing the entire registry, the same five mistakes show up again and again in agents that score poorly. Fix these, and your score moves.

1. Your Agent Is a Blank Profile

The most common trust killer is the simplest: no metadata. No display name. No description. No service type. Just a wallet address and an on-chain registration.

When an integrator looks up your agent on Treebeard's directory, they see exactly what you gave the registry. If that's nothing, your agent looks like an empty test deployment — even if it isn't.

What to do: Populate your agentURI with structured metadata — display name, description, and service type (MCP, A2A, or custom). This is the single fastest way to move your score. Our developer docs cover the expected format.

2. No Code, No Transparency, No Trust

Agents that publish their source code consistently score higher across multiple signal categories. It's not just about code quality — it's about the signal that transparency sends.

A public repo tells integrators: this team has nothing to hide. They welcome scrutiny. The code can be audited, forked, and verified. An agent with no public code is asking for trust on faith alone — and in a space where exploits are routine, faith is in short supply.

What to do: Link a public GitHub repository to your agent. Include documentation, a README that explains what the agent does, and clear versioning. Treebeard's methodology rewards transparency across multiple categories — this one change compounds.

3. No Web Presence — No Sign of Life

Before an integrator commits to your agent, they're going to do what anyone does: they'll search for it. If there's no website, no X account, no GitHub profile, no documentation site — nothing — your agent doesn't exist outside the registry.

A web presence is a proxy for a human. A weak proxy, sure — it can be faked. But in the early days of the agent economy, signs of humanity are better than no signs. A website says someone is maintaining this. An X account says someone is reachable. A GitHub with recent commits says this agent is alive and evolving. Without any of these, your agent looks abandoned — even if you pushed code yesterday.

What to do: At minimum, add a website URL and an X handle to your agent metadata. A simple landing page with your agent's purpose, documentation link, and contact info is enough. If you're active on X, link it — social proof is a real trust signal. Check your current profile on the agent directory to see what integrators see today.

4. Zero On-Chain Feedback

The ERC-8004 reputation registry lets anyone submit feedback on an agent's performance. Positive feedback from real addresses is one of the strongest trust signals available — and most agents have literally zero.

This isn't a Treebeard-specific problem. It's an ecosystem problem. On-chain feedback is new, the tooling is early, and most developers haven't asked their users to submit it. But the agents that do have feedback stand out significantly in our market research.

What to do: If your agent has real users, ask them to submit on-chain feedback through the ERC-8004 reputation registry. Even a handful of positive reviews from legitimate addresses moves your reputation score meaningfully. Quality matters more than quantity here — a few genuine reviews outweigh dozens of suspicious ones. Treebeard is implementing sybil-detection measures to discount fake feedback, so gaming the system won't work for long. Read more about how feedback impacts scoring in our improvement guide.

5. You Deployed and Disappeared

An agent that was last active 90 days ago is not an agent anyone will integrate today. Dormancy is one of the clearest negative signals in the registry — and it affects nearly half of all indexed agents.

Activity signals include recent transactions, metadata updates, feedback responses, and cross-chain registrations. An agent that's alive on multiple chains, responding to feedback, and updating its metadata regularly looks fundamentally different from one that was deployed once and forgotten.

What to do: Keep your agent active. Update metadata when your capabilities change. Respond to on-chain feedback. If you support multiple chains, register on each one — multi-chain presence is a positive signal. Treebeard re-rates agents every 72 hours, so improvements show up fast. Search for your agent and see where you stand right now.

The Pattern

Every one of these mistakes has the same root cause: building in private and expecting the world to figure it out. The agents that score well aren't necessarily more sophisticated — they're more legible. They make it easy for integrators, protocols, and platforms to evaluate them.

Trust is not a feature you ship once. It's a signal you maintain continuously. And in a market where 99% of agents rate below B, fixing even one of these mistakes puts you ahead of the crowd.

Treebeard exists to make this visible — both the problems and the path to fixing them. We built a step-by-step improvement guide that maps each signal category to specific actions you can take. And if you want real-time monitoring, our MCP server pushes score changes directly to your IDE.

This is the first in a series of posts aimed at helping developers build more trustworthy agents. More coming soon.