AI Trading Agents Settled $73M While DeFi Infrastructure Stayed Untested

Published by James Harris on June 11, 2026June 11, 2026

Autonomous AI systems collectively settled more than $73 million across roughly 176 million blockchain transactions between May 2025 and April 2026, and that volume is still small enough that the infrastructure holding it together has barely been stress-tested. The story is not that AI trading agents exist; it is that DeFi’s execution environment is genuinely hostile to them, and a new layer of middleware is forming around that problem.

Traditional trading bots operate on abstraction: submit an order via API, let the exchange handle custody, matching, and settlement. DeFi strips that abstraction away entirely. An agent executing a swap must check wallet balances, approve token spending, compare fragmented liquidity across multiple decentralized exchanges, estimate gas in real time, and model whether slippage will erase the trade’s edge before the transaction even lands. That operational surface is orders of magnitude wider than a centralized exchange API call, and it is where most automation breaks down. The needs here are not speculative; the 2024-2025 MEV and sandwich-attack data already showed that unsophisticated on-chain order routing is systematically extracted by searchers before retail or automated trades can clear cleanly.

The execution problem is not a product gap waiting to be filled; it is the reason most DeFi volume still flows through aggregators built for human-initiated trades rather than agent-driven ones.

Projects like Orbs Agentic are positioning themselves as decentralized execution infrastructure for this layer, offering liquidity routing across decentralized exchanges in a form that agents can call programmatically rather than relying on interfaces designed for wallets. The competitive angle here matters: if agent-driven volume scales, whoever owns the execution routing layer captures fee flow that currently goes to aggregators and liquidity protocols. Ethereum’s L2 ecosystem and Solana are both viable surfaces for this, but their different fee structures and finality guarantees mean infrastructure built for one does not port cleanly to the other. That fragmentation is a genuine moat for any team that solves multi-chain execution coherently rather than chain by chain.

The $73 million settlement figure will look either like an early signal or a ceiling depending on whether agent frameworks can resolve the gas estimation and slippage problem at scale. If they can, the more consequential shift is not trading performance but custody: agents that execute autonomously across wallets and smart contracts are also agents that can lose funds autonomously, and the risk management conventions for that scenario do not yet exist in any standardized form.