Code every day to get tasks done, mostly related to programming. Claude Code is great when you need to perform specific tasks. For example, if you’re assigned a specific task to do or if you’re working on a project.
However, Claude Code and other similar coding agents don’t work as well in proactive work. So basically, when you need to do something unprompted. This is where OpenClaw comes in very clutch.
OpenClaw agents are essentially coding agents that run 24/7. You can have them run in an application such as Slack or Telegram, and in essence, they work very similarly to a traditional coding agent. The main difference is that you access them via apps such as Slack, and they have some infrastructure around them, such as setting up cron jobs and so on.
In this article, I’ll discuss how you can set up a fleet of OpenClaw agents and have them effectively perform a lot of work for you proactively. This can help you a lot not only in coding tasks but also in any other task that you have to run within a business, for example, customer success management, finance management, sales, and so on.
Why set up OpenClaw agents
First of all, I also always like to cover the why you should care about a specific topic. And in this instance, it’s why you should care about setting up OpenClaw agents. The reason OpenClaw agents are so powerful is that they are agents that run 24/7, typically on a separate computer or in a cloud environment, so that they’re always available. This is in contrast to your Claude Code or other coding agents that are usually with you on the fly on your PC, but that you sometimes have to close, for example, if you’re traveling to and from work and so on.
As with most of the topics I cover, the main reason you should care about it is that it can save you a large amount of time and focus. You essentially can have OpenClaw Agents perform a large amount of the simpler tasks that you have to interact with on a daily basis so that you don’t have to perform the tasks yourself. This is great because, of course, it saves you a lot of time, but it also saves you a lot of focus because you don’t have to spend time thinking about the tasks at all. Which is, of course, great because it gives you greater focus, which you can spend on more important tasks.
How to set up OpenClaw agents
Now in this section I’ll start to cover how to set up OpenClaw agents. The topic of setting up your agent for the first time, just getting it to work, I’ll not spend a lot of time covering because I think it’s a pretty simple topic, and in essence, you can just ask a Claude Code session to set it up for you. The following prompt:
Set up an OpenClaw agent for me on called
. You can connect it to my Codex account with email .
Make sure to test the agent yourself and continue until you verify that it
works and come back to me once it works. Connect it to Slack. This will basically set up everything you need to have your initial setup of an OpenClaw agent. I recommend using Codex because you can use OpenClaw Agents with your Codex subscription. This is technically not allowed with Claude Code, and I find that OpenClaw agents work very effectively with Codex.
The more interesting topic, which I’ll spend more time on today, is how you can set up an OpenClaw agent as effectively as possible and how you can orchestrate a lot of them to work effectively together.
This is, of course, quite different from just setting up an agent and making it work to answer simple messages. I’ll cover this in a few subsections, which cover the most important points that you need to think about and implement to have OpenClaw agents work effectively.
- Knowledge base
- Cron jobs
- Teaching the agents tasks
Of course, it’s worth noting that the specifics of each agent will depend on what tasks the agent is performing and also your specific requirements and needs for that agent. So you of course need to generalize this to your own use cases and application areas.
Knowledge base
Having a knowledge base is one important topic that I think you should think about when setting up OpenClaw agents. I have two kinds of knowledge bases. One is the shared knowledge base that everyone in the company has access to, and that the coding agents consistently update whenever they have new knowledge added, new documents added, and so on.
I think this is super valuable, both to have all knowledge gathered in one location so that anyone, whether it’s a human or an agent, can find it on demand. It’s also worth knowing that when you set up this knowledge base, you should have a markdown file at the root that indexes all the content in the knowledge base, the structure of it, where to find certain information, and so on.
The second type of knowledge base that you should have is a bot personal knowledge base, which is basically knowledge the bot builds up over time. This is a knowledge base you can update with a daily cron job where the agent learns from its interactions that day.
If you have this setup, you’ll notice that the agents will improve over time and get better and better, which is super helpful for you when you want an effective fleet of OpenClaw agents. You should make sure to set this up for all of your agents so that all of them improve over time.
Cron jobs
Cron jobs are an incredibly important topic as well when it comes to OpenClaw agents, and it’s basically one of the most useful tools when it comes to making the OpenClaw agents proactive.
Cron jobs are essentially a schedule where you set up a certain task to run on a specific schedule.
This can be weekly, daily, hourly, or whatever schedule you prefer for the job to run. To help you understand better how to effectively use Cron jobs, I’ll give you some examples of Cron jobs that I’ve set up.
- Dreaming (Each agent dreams every night, learning from what it did or all the interactions it had that day to improve.)
- Daily QA bot I have a QA bot that goes through my application every day and reports the results in an HTML that it sends to me on Slack so that I have a full report coming into me every morning with potential bugs in the platform.
- Finance bot checking for invoices on a regular basis, and informing you when you need to pay them
In general, you should be very liberal with setting up cron jobs as they’re usually quite cheap to run. Of course, they require tokens from whichever coding agent you’re running, but each run doesn’t have to cost that much and can be super effective.
I try to be very liberal in setting up cron jobs with every recurring task that I can encounter. And I’d rather make too many than too few. If at a later point I notice that I have too many cron jobs or if a task is becoming too much noise, I can of course remove the cron job in the future.
Teaching the agent tasks
It’s a common misconception that OpenClaw agents will just work perfectly out of the box. This is not the case at all.
What I found is that in the beginning I need to spend quite a lot of time to manually teach each agent the specific tasks that I want them to perform. Say, for example, I have a finance agent that I want to go through invoices on a daily basis to see if everything is fine. I need to tell the agent exactly how to do that, what to look for, and so on.
Also, if you have a QA agent that goes through your application on a daily basis, you need to set it up. You need to give it a login so that it has credentials to access the platform, you need to inform it how to test the platform, and you need to tell it how to report its results. Should it be a Slack message? Should it immediately create PRs to fix any bugs that it finds? And so on. These are all things that you probably don’t even think about when setting up the OpenClaw agents, but that you need to teach the agents so that they’ll be able to perform the tasks effectively.
Conclusion
In this article, I discussed how you can orchestrate a fleet of OpenClaw agents. I discussed why you should care about setting it up. The most important argument is, of course, that you can save a lot of time by having OpenClaw bots proactively completing tasks for you, so you don’t have to spend time reactively completing them using coding agents such as Claude Code or Codex. I then continued discussing specific techniques and tips that you can use to make your OpenClaw agents more effective. I covered having knowledge bases to customize your agents and have them learn over time, setting up cron jobs to help with productivity, and teaching the agent how to complete specific tasks. I believe OpenClaw agents or similar concepts will become increasingly important in the future as we all have personal agents that help us automate large parts of both our work and personal life.
👉 My free eBook and Webinar:
🚀 10x Your Engineering with LLMs (Free 3-Day Email Course)
📚 Get my free Vision Language Models ebook
💻 My webinar on Vision Language Models
👉 Find me on socials:
💌 Substack