Automation is not automated
With the advent of AI agents it’s compelling to think we can simply throw tasks at them and they will do anything.
The reality is, to use them safely and responsibly there must be a lot of guardrails. They might be annoying, but they are necessary if we want to avoid disasters similar to Amazon outage due to AI agent dropping a database.
This is one of the reasons I haven’t jumped onto the OpenClaw bandwagon. I want to keep my server and projects intact.
I like that Claude Code asks me with “stupid” questions if I want to run this or that. And even after I allow it, I still have to tweak its output. Because it’s not 100% correct 100% of the time.