You don't need OpenClaw. All you need is /loop

Ok, I'm over-simplifying a little bit.

Unless you've not at all paid attention to anything remotely AI-related the last few weeks, you've heard about OpenClaw (or ClawdBot, as it was first called). The fastest-growing project on GitHub ever, and the one with the most stars ever. That's right: more stars than Linux! Many more, even. What took Linux decades has taken OpenClaw weeks.

That is not an affirmation of the quality of the project, though. However,

it is an attestation of the enormous hunger people have for agents!

Suddenly, it's almost like having your own Jarvis! (Just without the armor, of course.)

But with OpenClaw also came a lot of dangers. This has been thoroughly warned about and documented. Yes, there have been steps taken to alleviate some of the worst issues, but something with that much use is also a giant honeypot.

But what is it that OpenClaw gives you? At its core, it is basically three fundamental building blocks:
1. Memory
2. Tools
3. Scheduling

Memory is fairly easy to set up yourself. There are many ways to do this, but it is essential: without memory, the agent starts with a blank slate for every run. And, preferably, your agent memory should compound as your agents work (which you will experience as your agent learning).

There are lots of available tools out there, like MCP servers. Or web search. Or Bash commands. Without these, your agent can't do much.

And recently, Anthropic launched /loop. That was the final piece (scheduling) of the puzzle needed for building your own OpenClaw, and understanding what it does. Now, all you need to do is call /loop and your agent becomes proactive. Until OpenClaw, /loop and other ways of scheduling, agents were reactive: you ask them to do something, and they do it. You ask them a question, and they answer it. But with a scheduler like /loop it can run every morning at 09 and give you a summary of your plans for the day. It can connect to your second brain to get background info on the people you're meeting etc.

But it can also be set to do repeating, compounding tasks through the night. Andrej Karpathy used this principle in a project called autoresearch, which he then open-sourced. Tobi Lutke of Shopify took inspiration from this, let it run overnight on their Liquid project, and saw 53% faster parse+render and 61% fewer allocations over 93 commits.

You can try the same yourself without their technical skills by using /loop.

Sure, that's still not quite OpenClaw, but it shows how powerful something like the /loop is and how simple the concept of OpenClaw is.