Revisiting some of my priors about OpenClaw

I have to say, Peter Steinberg (aka lobster man) is absolutely brilliant. He was so ahead of his time with OpenClaw. I remember when he first tweeted about it last year, and when it turned into utter lobster madness at the beginning of the year.

Heck, I even ran OpenClaw for a bit, but I couldn't really see the value because I didn't want to maintain it while the lethal trifecta problem remains unsolved.

Categorically, this is a billion-dollar problem, perhaps more.

So here's the thing. I kind of want openclaw, but I want it managed. For almost a year now, I've been using Interaction Labs Poke, because if there's going to be a problem with the lethal trifecta, I want to make that problem someone else's as much as possible. I want them losing sleep over the lethal trifecta, not me. It's not full protection, but knowing there's a product team paid to think about these issues and resolve them at scale puts me at ease.

That is, until last week, when Poteto girl nerd-sniped me on X and, in full disclosure, hooked me up with enough credits to help me reevaluate my priors.

I am now using Grokbot full-time instead of Poke (which sucks), and I've revisited my priors about the utility of a personal assistance bot. In this post, I'm going to give a high-level overview of how I've automated some aspects of my life.

Okay, Grokbot is really weird, and I guess this applies to, like, even OpenAI's dots, even though I have not used it. One of the hardest things about it is how generic it is. Once you have it installed, it becomes a question of "now what?"

You need to be curious, have lots of imagination, and when you have an idea, you just need to ask for it. From that point forward, it starts to unfold like a pretty flower. If functionality is missing, just ask for it and ask it to turn into a skill.

If you're using Grok, install this skill pack first, and use it to create other bots. Every time you need a new bot, which is often, you should ask this bot to create a new bot.

Here are the bots I've defined in the last week, which I'm getting a lot of utility out of:

What you want to do is to create bots thinly. You don't want a fat bot. You want to create a bot per function. Think about the Unix command line, how composable that is. You want to create small bots that can call other small bots, then pipe them together in a function chain.

You can install software on your Grok bot. For example, I have Tailscale installed, and that's how I do all my tunneling.

If the lethal trifecta is ever solved at a fundamental level, I'll move all this stuff into a self-hosted Open Claw, but not before. I get the value of OpenClaw and open source, but fundamentally, I sleep much better at night knowing there's a product team whose arse is on the line, with staff and a funded security organization that loses sleep over these topics so I don't have to.