Agentic MEV Day [SBC]

A Field Guide to Agentic MEV: Emerging Problems and Methods

Agentic MEV has three tracks: AI for MEV, MEV-aware agent-network design, and potential new extraction created by how AI systems are built.

Presented by Jonathan Passerat-Palmbach

Efficiently using LLMs to find novel vulnerabilities in live protocols

LLMs can search deployed on-chain protocols for vulnerabilities, but the audit still has a finite budget.

Presented by Ivan Bogatyy

The Trading Agent Loop: How to build Claude Code for trading

A Claude Code-style trading loop turns an idea into a testable strategy, a backtest, and a cautious live deployment.

Presented by Michael Feng

Latency-optimized LLMs for trading

Lower-latency models can make financial-sentiment systems more useful for trading.

Presented by Sam Green

MuMev: test-time training for profitable discovery

MuMev asks whether an agent can rediscover profitable Ethereum strategies by writing and testing programs against historical blockchain state.

Presented by Luc Chartier

Learnings from Optimization Arena

Optimization Arena tests whether open challenges can produce useful quantitative-research and mechanism-design insights.

Presented by Dan Robinson

Information Liquidity

Calcifer Computing connects how language spreads and changes with a broader question: what would an information market look like?

Presented by Brandon Duderstadt

Cooperative General Intelligence

Cooperative general intelligence could preserve dispersed information through agents that make behavioral commitments.

Presented by Matt Liston

Information flow indecision markets

Decision markets use prices to help choose an outcome or organizational action.

Presented by Kollan House

NDAI & information flow control in DeFi

NDAI proposes that parties evaluate private information through agents while revealing only a narrowly defined decision.

Presented by Quintus Kilbourn

From Hardware to Harness

Agentic MEV spans the stack from hardware and compute infrastructure to the software harness agents use.

Presented by dmarz

Private Agent Memory

XTrace’s encrypted vector database is designed to keep queries and stored context hidden from the server.

Presented by Felix Meng

dstack Private Inference Gateway and Attestation Interoperability

Within the TEE threat model, private inference keeps model inputs and outputs confidential and provides attestation evidence about the workload and protected environment.

Presented by Hang Yin

The Agent Stack and Standards

The Agent Stack needs standards for identity, authorization, payment, verification, and communication.

Presented by Shashank Motepalli

Delegated World ID for agents using anonymous rate-limited credentials

Delegated World ID proposes anonymous, rate-limited agent credentials derived from a human user’s World ID.

Presented by Leyton Taylor

Compute Markets

A credible compute market needs delivery and a liquid spot price before financial products can price compute risk.

Presented by Waylon Jepsen & Sebastien La Duca