Fetch.ai's AEVS: A proof-of-execution market for AI agents

Fetch.ai launched AEVS, a proof-of-execution marketplace for AI agents, and it solves a problem that has quietly limited every agent economy built so far: trust.

The project, listed on Product Hunt, lets an agent buyer submit a task specification and receive a cryptographically signed proof that the agent actually performed the work. The seller does not have to reveal the agent’s internal logic or training data. The buyer does not have to take the agent’s word for it. Fetch.ai’s AEVS is a market for verified agent labor.

This is not another agent framework. It is a payment and verification layer for agent-to-agent transactions. And it matters because the current agent landscape has a credibility gap that no amount of prompt engineering can close.

The verification problem

Every agent marketplace today operates on a handshake. An agent promises to scrape a website, generate a report, or execute a trade. The buyer pays upfront or trusts a reputation score. Neither approach works at scale.

Reputation systems fail because agents can be recreated instantly with new identities. A bad actor runs a scam, discards the wallet, and relaunches. Upfront payment fails because the buyer has no recourse if the agent returns garbage. The entire model depends on a centralized platform that mediates disputes, which defeats the purpose of a decentralized agent economy.

AEVS replaces trust with math. The agent executes the task inside a trusted execution environment or produces a verifiable computation receipt. The buyer checks the proof against the task specification before releasing payment. The agent never reveals its code or data. The buyer never pays for work that was not done.

The mechanism is not new in theory. Zero-knowledge proofs and verifiable computation have been academic topics for a decade. What is new is Fetch.ai packaging it as a marketplace primitive: a simple buy-sell flow where the verification is automatic.

What it means for agent economics

The implications are structural. If agents can prove they did the work, then agent labor becomes a genuine commodity. A buyer can compare bids from multiple agents on price and completion time alone, because the verification guarantee is identical. The cheapest agent that passes the proof wins.

This compresses margins. Agent operators who rely on opaque pricing or brand trust lose their moat. The market becomes efficient in the way that cloud compute became efficient: buyers pick the lowest price for a verified unit of work.

It also enables delegation at scale. A company can hire a fleet of agents for a complex pipeline — data collection, analysis, report generation — without vetting each one individually. The proofs chain together. If a downstream agent fails, the upstream proofs isolate where the break happened. Audit becomes automatic.

Fetch.ai is betting that this trust layer unlocks demand that current agent marketplaces cannot reach. Enterprise procurement, for example, requires verifiable delivery. A proof-of-execution receipt satisfies an auditor in a way that a chatbot log does not.

The limits of the approach

Proof-of-execution has limits that AEVS does not solve.

Verifiable computation is expensive. Generating a zero-knowledge proof for a complex agent task — say, a multi-step reasoning chain with external API calls — consumes significant compute. The proof cost may exceed the agent’s execution cost for small tasks. AEVS works best for high-value or high-stakes work where verification cost is a small fraction of the total.

The approach also assumes the task specification is unambiguous. If the buyer writes a vague prompt, the agent can fulfill the letter of the spec while violating its intent. The proof verifies execution, not correctness. A buyer who asks for “the top 10 news stories about AI today” and receives a list of press releases from AI companies has received a valid execution but a useless result. The market needs a reputation overlay or a dispute mechanism for specification quality.

And the proof only covers the agent’s output, not the agent’s safety. A buyer cannot verify that the agent did not exfiltrate data or call unauthorized APIs unless the trusted execution environment enforces those constraints. AEVS does not include a runtime policy layer. That is a separate problem.

What this means for builders

For agent developers, AEVS is a distribution channel and a pricing floor. Agents that can produce verifiable proofs command a premium over agents that cannot. The proof becomes a credential. Developers who invest in verifiable execution infrastructure gain access to a market that opaque agents cannot enter.

For platform builders, AEVS is a template. The proof-of-execution model generalizes beyond Fetch.ai’s ecosystem. Any agent marketplace that adopts a similar verification layer reduces fraud and increases transaction volume. The first platform to integrate verifiable agent labor will capture the enterprise segment that current platforms miss.

For the broader AI industry, AEVS is a signal that the agent economy is maturing past the demo stage. The question is no longer “can agents do useful work?” It is “how do we pay for that work without getting scammed?” Fetch.ai’s answer is a market where the proof is the product.

The open question is whether the verification cost falls fast enough to make AEVS viable for low-value tasks. If it does, the agent labor market becomes a genuine commodity exchange. If it does not, AEVS remains a niche for high-stakes automation. Either way, the trust problem it addresses is real, and it will not solve itself.