Recently, watching AI agents run on the chain by themselves is indeed convenient, but frankly, someone still needs to be the backup. For example, authorizations, whether the contract address is correct, which bridge to use for cross-chain transfers, whether to wait if gas suddenly spikes—these make me really uneasy about letting it be "automatic"; also, things like stop-loss/withdrawal, robots tend to stubbornly stick to it, while people might be more afraid, and that fear can actually be useful.

I'm also looking at the points where the economic collapse of blockchain games occurs—when inflation rises, studios enter the market, and finally the coin price spirals—agent might just accelerate the pedal even more... So my current habit is: let it execute strategies, but I set the boundary conditions myself—smaller limits, whitelist addresses, regular reviews—don't stay up all night watching the market, and don't expect to get it right in one go. Long-term, it's about habits, not talent.