Trusting Agents With Capital: The Gearbox Playbook
summary
Claims a production-ready playbook: zero incidents, zero bad debt over 5 years, two years of pilots starting on EigenLayer, and a live demo that has an agent analyze ~100 docs in 5 minutes to recommend deploying $10k USDC while flagging an RLP token incident. Built tools: a proprietary indexer, graph memory with per-source credibility scores, an MCP server that generates human-readable verification links deployable to IPFS, and an on-chain bot APY harness capping per-transaction risk; structured product splits 90% passive yield / 10% agent-managed with on-chain stop-loss.
AI-written summary · about
chapters (37)
- 0:00Introduction
- 0:11Starting With the Number Zero
- 0:41Zero Incidents, Zero Bad Debt
- 1:07Two Years of Piloting Agent Experiments
- 1:45The Numbers We Are Facing
- 2:17The Gap Between Experimentation and Production
- 2:50The Vision for How DeFi Should Be Done
- 3:32DeFi 3.0: Ethereum as a Settlement Layer
- 4:09Why Peer-to-Peer Is Easier to Verify
- 4:45Why Autonomous Due Diligence Matters
- 5:16From Risk Curators to Curatorless Agents
- 5:48Why Autonomy Beats Manual Research
- 6:14The Gearbox Loop of Six Stages
- 6:39Why Shortlisting Is Needed
- 7:08Why Execution Must Be Safe
- 7:36How the Agent Collects and Analyzes Data
- 8:14Flagging an Insecure Pool
- 8:41Introducing the MCP Server
- 9:08Why You Never Sign What You Don't Understand
- 9:35The Human-in-the-Loop Verification Link
- 10:03Toward Autonomous Capital Management
- 10:47How Bot APY and Risk Budgets Work
- 11:12Why Capped Losses Are Easy to Hedge
- 11:38Protecting Against Leaked Agent Keys
- 12:07The Problem of Agent Memory
- 12:34Credibility Scores to Resist Manipulation
- 13:04Plugging In Institutional Due Diligence
- 13:37From Beating ML to Following a Manifest
- 14:05Why Different Goals Achieve Different Outcomes
- 14:48Mapping Interconnections Between Complex Assets
- 15:14Emergency Exits During Contagion
- 15:53Building a Structured Product on Gearbox
- 16:23Splitting Capital Into Passive and Active
- 16:51How the Agent Borrows and Competes
- 17:38The On-Chain Stop-Loss Harness
- 18:19Learning Like Robots Playing Go
- 18:43Closing
related talks
23mLeveraging AI & Layer 2 Blockchain Solutions
37mThe Production AI Playbook: Deploying Agents at Enterprise Scale
40mHow To Tame Your Agents?
26mContracts earn trust: API-first delivery for teams and for agents
15mThe 85% Beyond the Model
07h 47mAI Agents Conference 2026 - Day 5
21mBuilding Agents (the hard parts!)
07h 37mAI Agents Conference 2026 - Day 3