"Agents can be fully automated, and you're still watching the market?" — First, let me reply: they can run, but the blame still falls on the person😅.

Right now, I only ask AI to do three things related to on-chain interactions: scan pools, calculate slippage, and remind me to check before placing an order.
If there's anything that really needs human oversight, it's actually the most annoying steps:
Authorization (approve) — if given too much, it looks bad;
Cross-chain/bridging — when it gets stuck, Agents just "retry," but humans need to decide whether to stop;
And LP rebalancing — when on-chain congestion + MEV issues arise, automation can easily turn "peace of mind" into "heartbreak."
Recently, there's been talk about increasing taxes/relaxing and tightening regulations, and with deposit and withdrawal expectations changing, on-chain sentiment can flip instantly.
Agents can't understand that kind of intuition of "I don't want to gamble today"...
Anyway, I treat it like a coffee machine — it can grind beans and extract, but whether the final sip is bitter or not depends on me.