# Trusting Agents With Capital: The Gearbox Playbook | Mikhail Lazarev (Gearbox Protocol) at ETHConf

- Channel: [ETHGlobal](https://streameth.org/ethglobal)
- Date: 2026-07-09
- Duration: 18:50
- Topics: Gearbox Protocol, Mikhail Lazarev, ETHConf, AI agents, capital management, DeFi, agentic framework, autonomous vaults, curatorless, security, bot APY, risk budget, on-chain harness, stop-loss, MCP server, human in the loop, verification, EigenLayer, structured product, tokenization, RWAs, settlement layer, formal verification, peer to peer, credit lines, graph memory, credibility score, due diligence, risk curators, indexer, USDC, insurance, contagion, emergency exit, institutional
- Watch: https://streameth.org/watch/yt-HcD37L1mVxI
- YouTube: https://www.youtube.com/watch?v=HcD37L1mVxI

## Description

In this talk, Mikhail Lazarev from Gearbox Protocol shares his playbook for trusting AI agents with capital, opening on the number zero: Gearbox has had zero security incidents in five years and zero bad debt, and zero could also be how much money is left if you fail to secure agents properly. Drawing on two years of piloting experiments starting with prototypes on EigenLayer in 2024, he lays out the Gearbox agentic framework and how anyone can set up a curatorless autonomous AI vault today. He frames the stakes with numbers: BCG projects $19 trillion tokenized on chain by 2033, and while 62% of entities are experimenting with AI per McKinsey, only 10% use it in production, a huge gap. Referencing a recent tweet from a well-known security figure warning that AI is now so good at finding vulnerabilities that DeFi feels unsafe, he argues DeFi 3.0 needs two shifts: moving from complicated monolithic smart contract systems to using Ethereum as a settlement layer with simple one-to-one peer-to-peer credit lines that are easy to formally verify, and moving from human risk curators to curatorless due diligence performed by agents that research deeply, respond fast to market changes, scale, and never sleep.

Mikhail walks through the Gearbox loop of six stages (discover, analyze, shortlist, propose, execute, and monitor), noting they built their own indexer to solve the lack of proven data sources. He demos an agent analyzing around 100 documents in five minutes to recommend how to deploy 10k of USDC, flagging an insecure pool with an RLP token incident, and introduces an MCP server so the loop runs in any coding agent with a human-in-the-loop verification link (deployable to IPFS) so users never sign what they do not understand. He explains bot APY, an on-chain harness using Gearbox's battle tested total-value computation to cap an agent's risk budget per transaction, making losses easy to hedge with insurance and protecting against leaked agent keys. He covers graph memory with per-source credibility scores to resist toxic tweet manipulation and plug in institutional due diligence, mapping interconnections between complex assets to enable emergency exits during contagion. He closes with a structured product where 90% sits in a yield pool while an agent borrows against it and competes to outperform, backstopped by an on-chain stop-loss.

00:00 Introduction
00:11 Starting With the Number Zero
00:41 Zero Incidents, Zero Bad Debt
01:07 Two Years of Piloting Agent Experiments
01:45 The Numbers We Are Facing
02:17 The Gap Between Experimentation and Production
02:50 The Vision for How DeFi Should Be Done
03:32 DeFi 3.0: Ethereum as a Settlement Layer
04:09 Why Peer-to-Peer Is Easier to Verify
04:45 Why Autonomous Due Diligence Matters
05:16 From Risk Curators to Curatorless Agents
05:48 Why Autonomy Beats Manual Research
06:14 The Gearbox Loop of Six Stages
06:39 Why Shortlisting Is Needed
07:08 Why Execution Must Be Safe
07:15 Demo: Deploying 10k of USDC
07:36 How the Agent Collects and Analyzes Data
08:14 Flagging an Insecure Pool
08:41 Introducing the MCP Server
09:08 Why You Never Sign What You Don't Understand
09:35 The Human-in-the-Loop Verification Link
10:03 Toward Autonomous Capital Management
10:47 How Bot APY and Risk Budgets Work
11:12 Why Capped Losses Are Easy to Hedge
11:38 Protecting Against Leaked Agent Keys
12:07 The Problem of Agent Memory
12:34 Credibility Scores to Resist Manipulation
13:04 Plugging In Institutional Due Diligence
13:37 From Beating ML to Following a Manifest
14:05 Why Different Goals Achieve Different Outcomes
14:48 Mapping Interconnections Between Complex Assets
15:14 Emergency Exits During Contagion
15:53 Building a Structured Product on Gearbox
16:23 Splitting Capital Into Passive and Active
16:51 How the Agent Borrows and Competes
17:38 The On-Chain Stop-Loss Harness
18:19 Learning Like Robots Playing Go
18:43 Closing

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _  

*ETHConf 2026*
ETHConf is a 3 day event for founders, industry leaders, and builders who are excited about the possibilities of building on top of Ethereum.

Connect with 2,000+ top innovators in crypto, finance, technology, and policy at our inaugural three-day event packed with showcases, demos, partnerships, and conversations shaping the future of the global economy.

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _  

✅ *Follow ETHConf*
X: https://x.com/ethconf
Website: https://ethconf.com

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _  

🎟️ Get your early bird tickets for ETHConf 2027: https://ethconf.com/2027#tickets
🎤 View the full ETHConf 2026 Speaker Schedule: https://ethconf.com/schedule

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _