20-30% better data efficiency, and in some cases 50 causal annotations getting manipulation tasks over the hump, is the line DeFi people should care about: lower training cost is how agent wallets graduate from chat wrappers into systems that can test actions before touching real capital. A causal model that can reason over interventions maps cleanly onto Aave/Compound liquidation cascades, Uniswap routing, Chainlink oracle drift, and governance simulations where correlation-only bots get chopped up. The check is whether Aether publishes benchmarks outside selected robotics tasks; until then, the investable angle is agent infra research, not tradable AI x crypto alpha.

Top comment by @Benthic