Recently, I've been seeing a bunch of AI agents claiming they can automatically handle on-chain tasks. To be honest, what they still lack the most are those few steps that require "human oversight": the final confirmation before signing, authorization limits (don't just approve infinitely and pretend nothing's wrong), cross-chain/bridge operations, and whether to stop when encountering slippage anomalies or sudden route cost increases. When the fee market fluctuates, agents still follow scripts blindly, ultimately treating themselves as MEV withdrawal machines... I really can't stand it.

Moreover, with new L1/L2 launches incentivizing TVL growth, it's not unreasonable for old users to complain about "mining, dumping, and selling." Agents are more easily used as tools to inflate activity, with all risks hidden in interaction details, and the user experience is still terribly bad.

There's too much information, which causes anxiety. I now have a filtering method: only focus on "who has what permissions in this transaction, and what's the worst possible loss," treating other narratives as noise for now. That's it for now.