A tool built on top of coding agents like Claude Code. It allows you to have a code agent working 24×7, working diligently and diligently to solve tasks. I've set up a lot of OpenClaw scenarios and learned a few good things about using it effectively. I've also discussed it a lot with my colleagues who work with OpenClaw agents every day, and in this article, I'll share some tips and tricks I've learned on how to get the most out of OpenClaw and some mistakes I've made, which I'll tell you how to avoid.
Why set up OpenClaw
The main reason you should set up OpenClaw is that it can make you successful as a developer. Where previously you had to run everything in Claude Code and be on your computer and ready to work at all times, OpenClaude can be run from a different computer and accessed from anywhere using applications such as Telegram or Slack. This makes it incredibly easy to work with a code agent like Claude Code, and you can contact them from anywhere.
In addition, you can easily set up cron jobs and skills, making the agent run code periodically and always remember to do so. And it can have skills that it loads when needed to better perform certain tasks.
All in all, OpenClaw just makes your coding agents a better assistant. It makes it more available and better able to perform tasks.
Error 1: Not working on Docker
The first mistake I made was not using OpenClaw in Docker containers. There are many reasons why you should use OpenClaw agents in Docker containers, and I will list a few of them here.
- It is very secure. Your agent cannot access everything on your computer; it can only access what is available in the Docker image.
- It is very easy to make backup copies of your agent and move them anywhere, as you can simply download the Docker image and use it elsewhere. This works because a Docker image is a completely separate container that can be run by itself.
- If you are running multiple agents on the same computer, it is better to separate them so that there is no overlap between your agents.
Overall, there is no real reason not to run on Docker. It's also very easy to set up using OpenClaw on Docker as you can just ask your coding agent to set everything up for you. In fact, you don't have to do anything yourself, and coding agents are extremely good at setting up a Docker system. When I did this for myself, I actually didn't have to do anything other than tell the model to set up OpenClaw on Docker, and it used it without a problem.
Mistake 2: Not giving the agent proper training
The second mistake is not giving your agent the proper training and setup help they need to do it right. When I set up my first agent, I spent more than ten minutes explaining what it had to do, gave it the necessary permissions, and hoped that would be enough.
Turns out that's not how you do it at all. What ended up happening was that my agent could not really do any of the tasks he was supposed to do because he had not received specific training on how to do those tasks. I, for example, gave my agent access to AWS without telling them how to access AWS, how to use it, how to communicate with people through Slack, and so on.
What ended up happening to me was that the agent started contacting people on Slack in messages they shouldn't have responded to. And when it was marked directly, it didn't know exactly what to do in those situations.
To solve this problem, you need to give your agent very specific training and tell them what they should do, what they shouldn't do, and how they should perform the tasks you ask them to do.
For example, if you give it access to AWS and allow it to interact with people through Slack, you should:
- Refer to the AWS documentation to know exactly how to use it and not make incorrect API calls or SDK calls.
- Explain to the agent which messages to respond to and which messages to not respond to, which are appropriate, basically.
- Explain to it the different questions people may ask and how it should answer those questions. For example, if someone asks about a specific customer, it should look up that customer in the customer table, look for different conditions that match this customer, and then ask the user clarifying questions.
Mistake 3: Not giving your agent enough permissions
Mistake number three is when you set up your agent correctly but you haven't given it enough permissions to do what it needs to do. For example, if you asked your agent to do a bunch of AWS tasks but you don't give it enough access, for example, it can only access DynamoDB but not S3 completely, it is very difficult for the agent to do the job.
When you set up an agent, you should think of it as if it were a person. If you've given the intern a bunch of tasks to do, but haven't given the intern the AWS permissions needed to do the job, it can be very difficult for the intern to know what to do.
The employee would not be able to ask for permissions, for example, or he might not know because he has never faced this situation before. Or it may think that it has to find the object itself, when in fact, you have to give it the permissions it needs.
Therefore, you should do the following when setting up the agent.
- Think carefully about everything an agent needs to do and make sure you have access to all the right resources. And if you don't grant access to certain services, be sure to let the agent know that they don't have access to this and how to respond to people who ask questions that require such access.
- Give the agent access to everything he might need, of course, within security constraints. This probably includes read access to almost everything you have, simply because read access is indestructible.
- Monitor the performance of the agent, especially at the beginning of its setup. If you see that the agent is struggling with certain tasks, you should help the agent by telling him how to solve such tasks. You must also grant or revoke access that the agent needs or does not need.
Overall, it all comes down to monitoring your agent and making sure it's working as expected.
The conclusion
In this article, I have discussed three common mistakes that are made when setting up OpenClaw and that I have made myself when setting up OpenClaw agents. These mistakes greatly reduce the performance of OpenClaw, so I highly recommend following all the tips I gave in this article and avoid the three mistakes I listed. Overall, though, it all comes down to monitoring your OpenFlow agent and helping it where you can see that the agent is struggling. If the agent is struggling with certain tasks, it is probably not an agent problem, but a setup problem or user error. Therefore, you should monitor your agent and make sure that it is effective in the tasks that you ask it to perform.
👉 My Free eBook and Webinar:
🚀 10x Your Engineering with LLMs (Free 3-Day Email Course)
📚 Get my free ebook Vision Language Models
💻 My webinar on Vision Language Models
👉 Find me on social media:
💌 Stack
🐦 X / Twitter
