OpenClaw and the shift to AI employees
Yesterday a post came across my X timeline about a guide to “OpenClaw”. The concept is pretty interesting, and it made me feel that the way we use AI is about to shift.
We’re going to move from just asking a chatbot questions to hiring “AI employees” that have their own schedule and take care of admin work.
An assistant with a heartbeat
Until now, we’ve used AI only when we need it. In OpenClaw, the AI has its own automatic work rhythm, called a Heartbeat.
Every 30 minutes it can “wake up”, check the schedule and do routine tasks without being told. It’s like having an assistant who’s already at their desk every morning, without you having to wake them up.
A team, not one AI
We don’t build one AI to do everything. We design a “team” with specific roles. The rules live in three simple files:
- SOUL: the SOPs and how it should behave.
- USER: our preferences, so it doesn’t keep asking the same things.
- TOOLS: the skills it’s allowed to use. For example, it may read email but isn’t allowed to reply.
Say we build a Personal Assistant. Every Friday afternoon it checks the calendar and sends a message on Telegram to coordinate the weekend logistics.
Or a Sales Assistant. As soon as a new client comes in, it automatically researches their background and sends us a briefing 30 minutes before the meeting.
Running it safely
This is the part that put my mind at ease. The AI doesn’t run on the servers of some outside company where we don’t know what’s going on inside. It runs on our own machine.
You can use an old laptop you don’t use anymore, or rent your own VPS. So it’s not on your everyday laptop, and your sensitive data stays safe.
Because it can act on its own, the risks are real. The ground rule: don’t give it write permission right away. Start with read-only access first.
This is where our ego gets tested. Are we brave enough to delegate small things and accept that the results might not be 100% perfect at first?
From prompter to systems manager
Going forward, our mindset has to shift, slowly. From being busy as a “Prompter”, hunting for just the right words, to being a “Systems Manager”.
Just start small. Build one agent to take care of one daily problem, the most boring and repetitive one you have.
The concept sounds a bit technical at first, but the goal is actually very human: to give us more time and energy to think about the strategic things AI will never be able to do.
It’s really interesting, and something I want to learn slowly.
Here’s the guide: OpenClaw: The complete guide to building, training, and living with your personal AI agent, on Lenny’s Newsletter.