Recently, I've seen everyone hyping up AI Agents automatically on-chain doing work, and I'm quite excited but also a bit scared. Checking balances, scanning large transfers, helping me monitor address activity—letting it handle these is no problem; but when it comes to actually signing, honestly, a human still needs to be the backup, at least to confirm: who exactly am I giving to this time, how much, is the permission too broad?

Contract interactions are even more complicated—approvals, authorizations, revocations—one wrong move and it could turn into a "long-term subscription," and I don't want to rely on luck.

And on-chain isn't a webpage; a quick hiccup isn't just a matter of "refresh and retry." Sometimes you have to wait in line for confirmation, and the Agent might think it failed and try a few more times... fees and risks stack up together.
The previous blockchain game economic collapse was quite similar—inflation + studio + coin price spiral, the more scripts run, the faster it dies.
Anyway, my current principle is: only let it do observation and reminders automatically; for key actions, I still click manually.