For the past six months, I've used a bunch of agent frameworks and various modified "AI personal assistants," until yesterday when I spent the whole day tinkering with Clawdbot—and I think I've gotten a glimpse of the prototype of an AI OS. It's a "Frankenstein's monster": multi-model CLI integration + Skills system + local web-based management + multi-platform connectors simultaneously supporting WhatsApp, Telegram, iMessage, Slack... It can understand voice messages, has short/medium/long memory, can help you manage your calendar, read emails, run scheduled tasks, and even program... Even non-programmers can use it.

After using it for a while, I started to resent Apple's lack of support. Apple could easily have done this themselves, perhaps even better understanding your needs—it wouldn't be complicated, it would be secure, and they could directly control AI within iMessage. Unfortunately, they didn't. Essentially, it's a system-level AI scheduler: • Step-by-step permission granting • On-demand skill loading • Persistent memory for clearer intent • Multi-channel input/output It's only missing operating system-level integration to be a true AI OS.

We will continue to observe the evolution of these products. If you are also looking for the prototype of an "operating system in the AI era," you can check out docs.clawd.bot/start/getting-s...… An interesting era.

Using it gave me a feeling of having tainted my computer; it was both pleasurable and I wanted to reject it, a bittersweet experience...

Claude's speed was too fast to catch up with.

That feeling, I guess.

I'm scared, I'm scared.