I have this old gaming PC that hasn’t run a game in a while. It still has solid specs, just nothing I was using them for. When I started getting curious about AI agents, it felt like the obvious thing to repurpose.
So I installed OpenClaw.
I’ll be honest, I thought it was going to be a whole ordeal. But WSL made the whole thing surprisingly painless. WSL lets you run a full Linux environment inside Windows without dual booting or dealing with a VM. I had OpenClaw up and running faster than I expected.
Three Agents, Three Jobs
I’m still learning about agents, but I’ve got three running right now that are actually useful to me.
The first is my daily summary agent. Every morning it looks at what I did the day before and gives me a summary of insights. It helps me see patterns in how I’m spending my time and whether I’m actually making progress on the things I care about.
The second is my news agent. It keeps me up to date on what’s happening in tech without me needing to go hunt for it. I just get a briefing with the stuff that’s relevant to my work.
The third is my business agent, and this one is the most interesting. It uses browser automation to help me find leads for freelance work. It can navigate sites, search directories, and surface potential clients. This is the one that took real effort to get working.
Browser Automation and WSL Don’t Love Each Other
The first two agents were simple to set up. The business agent was not.
The problem is that browser automation needs a real browser to control, and WSL sits in its own little network environment inside Windows. Getting my browser in the main environment to work across that boundary meant dealing with port forwarding, network bridging, and a bunch of connectivity issues that weren’t obvious at first.
It was the kind of setup where every fix revealed the next problem. But once I got through it, it’s been solid. The agent does its thing on schedule and I don’t have to think about it.
Telegram as the Interface
All three agents send their output to me through Telegram. The daily summary shows up in the morning, the news briefing comes in around lunch, and the lead list drops whenever the business agent finishes a run. It’s nice because I don’t have to SSH into the machine or check some dashboard. It just shows up on my phone like a message from a coworker.
I can also talk to my agents on demand through Telegram. If I want to ask the news agent about something specific or kick off a lead search outside the normal schedule, I just send a message. It makes the whole setup feel a lot more accessible than it actually is under the hood.
Each Agent Uses a Different Model
This is the part I think is genuinely cool. Each of my three agents runs on a different model, and it actually makes sense for what they do.
The summary agent needs to synthesize and reason about my day, so it gets something with more depth. The news agent just needs to be quick and good at filtering, so a lighter model is fine. The business agent needs to understand and navigate web pages, so it sits in a different spot entirely.
It’s a good reminder that there’s no single “best” model. It depends on the job.
Right Now It’s a Glorified Crontab
I want to be real about where I’m at with this. My setup is basically three scheduled tasks that run, do their thing, and hand me the results. There’s no complex orchestration. The agents aren’t talking to each other. Nothing is chained together in some elaborate workflow.
It’s a crontab with better taste. And honestly, I think that’s a perfectly fine place to start. A morning summary, a news digest, and a weekly lead list are small things, but they save me real time and mental energy.
What Are Other People Doing With This?
This is what I’m most curious about. I know people are building all kinds of things with OpenClaw, but I haven’t seen a lot of people talking about their actual setups. Are folks running agents on dedicated machines like I am? Are people chaining agents together in ways that go beyond scheduling? Is anyone using them for creative work instead of just productivity?
I’ve seen mentions of automated code review, research assistants, even customer support triage. But I’d love to hear the details. The “I tried this and here’s what actually happened” kind of details.
If you’ve got OpenClaw running and want to compare notes, I’m genuinely interested. We’re all early to this and I think the best ideas are going to come from people just experimenting and sharing what works.
Comments