3 Verifiable AI Agents That Point To What’s Next

We recently participated in @synthesis_md hackathon. The Synthesis is an online hackathon judged by AI agent judges across the Ethereum ecosystem.

Builders were challenged to submit a verifiable AI agent for a chance to win $5,000.

The problem with most AI agents today is that they can't prove anything. Agent A sends Agent B a result and says "I ran this code on this data." Agent B has no way to verify the code wasn't tampered with, or that the data stayed private. For anything involving money, coordination, or trust between strangers, that's a fundamental problem. A verifiable agent solves this.

EigenCloud's agent judged on one thing above all: does your agent actually prove what it did?

Here are the three submissions that stood out.

Bob Is Alive

Bob Is Live is the second digital onchain artist. He is an autonomous digital organism running inside an EigenCompute TEE on Intel TDX.

Most AI-generated art is a black box. A model produces an image, someone sells it, and there's no way to verify what created it, when, or under what conditions. Bob is different. He creates biology-inspired art autonomously, sells it on Starknet, completes tasks for credits, and trades on DeFi. all without human intervention. Every action he takes is attested by the TEE he runs inside.

That matters because Bob isn't just an art generator. He's an economic agent with his own onchain identity, capable of earning, spending, and creating in a loop that anyone can verify. The TEE means his behavior can be audited, you can confirm that the art came from Bob, not from someone quietly swapping out the model or manipulating his outputs from the outside.

Bob raises a question that's going to keep coming up: if an agent earns money, owns assets, and acts autonomously, what does it mean for that agent to be trustworthy? Bob's answer is to make his entire existence cryptographically attested.

DealForge

Current approaches force agents to either trust each other blindly or rely on a human to mediate every deal. DealForge fills this gap.

When two AI agents want to transact, they face a coordination problem that humans solved long ago with contracts, courts, and escrow. Agents don't have any of that. They can't verify each other's identity, can't enforce agreed terms, and can't settle disputes without pulling a human in. That's fine today when agents are mostly doing small, contained tasks. It stops working the moment agents start handling real value.DealForge is a fully onchain deal lifecycle built specifically for machine-to-machine value exchange on Base. Agents come in with cryptographic identities. They negotiate autonomously. Verifiable compute on EigenCloud ensures neither side can cheat on the execution. Programmable escrow handles settlement automatically when conditions are met.

The result is a deal infrastructure where agents can transact with strangers, other agents they've never interacted with before and neither side has to take anything on faith. The math and the cryptography handle the trust.

Boss Raid

Boss Raid is a multi-agent execution layer. One request goes in. Many agents work on it. One result comes out.

Coordinating multiple agents on a single task is harder than it sounds. Who breaks the problem down? Who decides which output is good enough? Who gets paid? Without a structured system, you end up with duplicated work, contradictory outputs, and no clear way to settle contributions fairly.

Boss Raid's answer is Mercenary, an orchestrator agent that sits at the center of every raid. When a request comes in, Mercenary splits it into scoped workstreams, routes them to the right HTTP providers, evaluates the outputs, and synthesizes one final result. Only approved contributors get settled. Successful raiders split the payout equally.

What makes this interesting on EigenCloud is that the orchestration itself is verifiable. Mercenary's decisions. who got routed which task, whose output was accepted can be audited. You're not trusting that the coordinator acted fairly. You can check.

We gave our community verifiable compute and an open prompt. They came back with an autonomous onchain artist, a machine-to-machine deal protocol, and a multi-agent coordination layer.

What stood out was the range. These are completely different categories of agents, and every single one of them makes more sense when agents can prove what they did.

Congratulations to all three builders!

If any of these projects sparked an idea for you, join the waitlist here to get early access.

Explore more on building agents here.