I don’t like being caught up in a hype cycle—I really don’t—but I have to talk about ClawdBot. It has impressed me more than any other AI branded product I’ve encountered thus far and it’s done so in a short amount of time. The closest analog for me was the first time I used Claude Code. I’m still in the honeymoon phase and have yet to experience a catastrophic fumble of my personal data, so take my excitement for what it is.

ClawdBot has a level of autonomy and judgement that make interacting with it feel like working with a rational, thinking actor instead of babysitting a powerful tool that lacks common sense. It’s been broadly competent out of the box in a way LLMs haven’t for me. I didn’t spend time combing through a skills directory, writing detailed instructions and context documents, or setting up projects to limit its scope so it stays on the right track.

Instead, I’ve asked it for help and in return it has tried its best to do those things. Sometimes it needs things from me. When it does, it asks and/or directs me where I need to go. Then it goes back to doing what it was doing, which was helping me. I know that sounds simple but that’s the point. I haven’t had to work hard¹ to get it to work for me.

Example convo below

A lot happened in that exchange. Here’s what impressed me.

• it located and modified its own configuration and then restarted a service that is essential to clawdbot‘s functioning
• it browsed the local file system, parsed files, and found something relevant based on the contents of the file
• in addition to above, it noticed an incomplete task that has a due date soon and proactively raised it
• it advised and set up an API integration on its own and then returned to the original request it couldn’t complete beforehand
• troubleshot an issue with a service (one it set up on its own previously) when I suspected something wasn’t working right

Next example (fyi, I truncated the json config file)

My notes

• it did indeed set the default Claude model I asked it to earlier
• it ran through a series of troubleshooting steps, checking the local machine its running on and collected necessary data from me along the way to arrive at a solution
• it fixed the problem and recorded solution-relevant details to avoid the it in the future
• it also decided to record the aliases I use for my lights for smoother interactions in the future

Later on, I asked it about that clawdhub issue it called out during the homebridge set up and it rolled through troubleshooting on its own (after taking a moment to correct its dumb human) without any additional input from me.

Maybe Claude Cowork is just as good but I gave up on it because it’s confined to a single machine (my personal laptop) that I only use for an hour or two a day at best. I can access ClawdBot on every device I have, 24/7. Could I have achieved all this with my own configuration of services and integrations via n8n? I think so but I haven’t had the time to learn how all that works.

It’s not really what ClawdBot does that is most impressive to me; it’s the way it does what it does. The UX is smoother than similar tools I’ve worked with and when it comes to the mainstream that means a lot more than it does when compared to early adopters. Especially with new technologies that haven’t saturated their market. This is how I expected New Siri to work when Apple pretended it existed almost two years ago. This is what Gemini should be doing now. clawdbot‘s competency tells me that magical feeling sci-fi personal assistants for all are much closer than I thought.

¹except for set up, which, if you don’t do locally, is onerous