Seeing someone say "AI agents will handle all on-chain interactions automatically in the future," I find it hard to believe... Frankly, who takes the blame for signing? Even the smartest agent can be tricked by prompt injections or baited with "seemingly normal" call data in phishing contracts. People still need to review permissions, limits, and whether unlimited approval is possible at the end. And for cross-chain/routing, once the wrong path is chosen and slippage explodes, there's no undo button on the chain. Recently, everyone is watching staking unlocks and token unlock calendars, shouting about selling pressure, but I'm more concerned about whether agents will reckless leverage or chase high during volatility. Risk control parameters need human oversight, at least a confirmation step. As for "just leave it all to the strategy"... forget it, no matter how well written the strategy is, it can't withstand a small bug or a re-entrancy attack.