Introducing Verifiable Agents on EigenLayer

EigenLayer proposes "Level 1 Agents," which use verifiable services built as AVSs for tool calls, because running agents fully onchain is impractical. The post says LLM attention memory grows quadratically with prompt length, becoming a bottleneck past about 100k tokens, and that inference cannot practically run inside a smart contract. Its claimed basis for verifiability is that AVS operators validate outputs because they risk being slashed and losing future revenue. Sreeram Kannan wrote in a January 5, 2025 post that agents should run as AVSs to achieve unstoppability.