OpenClaw in most cases also going to use the very same dedicated tools, maybe variation of those tools dumbed down for LLM.

Almost every time I have an idea for AI Agent, I end up just making a script/binary that does the same, but so much faster that adding AI to it feels silly.

Recently I made a tool router that runs locally for such tools. Some tools have no arguments at all. Claude created a quick overlay where I can text/speak, and it will do tool call, without me asking for it, Claude added 4 buttons next to text input that bypass agent and just do a "tool call". I barely use text-to-command because those 4 buttons cover 9/10 of my use cases.

At this point I'm trying to come up with tools to add to it, so it's actually useful as an agent. Almost everything ends up being a cronjob or webhook triggered thing instead.

My experience also. I could manually connect my Obsidian notes to my AI, sure, but what I did instead was writing "Obsidian just released a CLI headless sync tool, install it so we can use it" and in a minute it came back with "Ok, everything installed, I just need your login and password."

Dangerous? Yes, very, but it truly feels like living in the future. Surprisingly, it's even more fun that sci-fi movies made me think this would be.