A very popular topic is coding agents lately. Loops are a concept where you have an agent that acts autonomously and you have an agent act autonomously.
This is in contrast to how we used coding agents before, where you fire up a new recording agent, have the coding agent execute something, validate its results, and so on until you're done.
Loops are incredibly powerful because they allow you to take a back seat as a human and multitask. This is because you don't have to follow the agent closely, and it works independently by itself and can complete the task in the end.
In this article, I'll discuss why you should work with loops when working with coding agents and how you can do it. I'll cover some techniques that you should actively use when working with coding agents to effectively stop loops and make your coding agents as efficient as possible.
Why use loops and coding agents?
First, let's cover why you should use loops when using coding agents. The simple answer is that loops allow you to do more work.
The reason for this is that when you set up a loop, the agent is able to automatically complete end-to-end tasks, which frees up your time as a human, allowing you to do more work.
You can imagine these two situations:
Scenario 1: You spin up agent A, make a plan for it, and then ask it to go on a mission. Then you spin agent B, and before you finish scheduling the task with agent B, agent A asks questions or tells you that some task has been completed and needs you to confirm it. You decide to get rid of agent B, and right after you're done there, you go back to agent A. Contact him, make him stay active, and before you're done there, agent B needs to be added to you again. Then you continue like this, and you can only communicate with two agents at the same time, in fact, you complete two tasks at the same time.
Scenario 2: You spin up Agent A and have it use a loop to verify its work. Agent A will then start working on a task, and you tell it to get back to you only when it has completed that task. Once you have finished commanding Agent A, you can start doing the same for Agent B, setting a goal and making it run in a loop. In this case, Agent A doesn't bother you for more because it has a self-validation loop and doesn't need your input in the same way. Therefore, you can continue to set tasks for agents C, D, E, and so on, until agent A completes its task.
The bottom line here is that scenario 2 allows you to integrate more agents and complete more tasks, which is, of course, very important because it allows you to do more work at once.
How to work with loops
Now, the big question, of course, is how do you work in loops? There are many different ways to do it, but I'll cover a simple method that you can start using right away.
This process is to use the /goal command with Claude Code or Codex.
/goal I, for example, say:
/goal Implement everything I asked for. Verify it end to end by clicking
through the browser using the Playwright MCP. It's not acceptable to test
the application only through integration tests. You need to actually click
around the app. Continue like this until it works. Fix any issues if you
encounter them then do an end to end test again. Run Codex exec and run
the review skill with Codex and make him approve it and iterate until
Codex has approved it. When Codex has approved it, come to me and tell
me which servers I can test it on and exactly how to test it.Basically, what it does/goal is to use a hook. This hook is triggered every time a Claude Code or Codex completes its task, and makes the scripting agent think whether it has completed the task it wrote under/in the target or not. When it's done, it returns to you and informs you of its work. If it thinks it is not finished, it will continue to work towards that goal until it is happy.
It's actually a way to make coding agents keep working until you reach your goal, even if they believe it's completely unattainable. A very powerful way to make coding agents work longer, and thus work independently and complete more work.
Also, notice in the notification that I give the agent a lot of other information about how to verify their work. This is a topic I'll cover in the next section, as it's incredibly important if you want to get the most out of the /goal command.
The approach/goal worked well
In this section, I will cover how to make the command/goal work. Of course, you can just give the agent a goal, but if you are not specific about what goal you are setting and how the agent can ensure the goal, you will not get good results.
You need to give the agent a way to validate its work and reduce the chance of bugs. I do this in two main ways:
- Tell the agent to verify their work end-to-end using Playwright MCP and browser integration.
- Have the agent use Codex Exec to review the code it provides to reduce the chance of bugs
I'll cover the first point: I used Playwright MCP, but you can use any browser interface you want. I like Playwright because it works very well, and the agent is able to complete all their work. Telling the agent to verify its work from the end works very well because the agent goes to my browser or app, takes screenshots, and verifies that everything is working as intended. This is an incredibly powerful and really simple change you can make to at least double the performance of your coding agents.
If you're not working with the UI, you can tell the agent to make API calls and check databases or logs to see if things are working as expected. The whole point is that not only does the coding agent verify its work by looking at the code. You run the code and have it validate the output and results of the code.
Second, I tend to use Claude Code as my main driver when developing code, but I've found that you can still get a lot of value from Codex.
I use Codex to provide my agents, my Claude Code agents, with updates. Whenever they publish code or want to publish code and get code to dev, I have Codex review the code and approve it before Claude is allowed to compile it into dev. This saved me a lot of bugs, it's hard to understand.
There is also a big difference in my experience with Claude's Code updating the code and the Codex. If I use Codex Claude as my primary driver to use the code, I definitely want to use Codex to verify the code later and not Code Claude. This is because the Codex just happens to be able to find some bugs. I create a recurring setup where Claude Code runs the code, has the Codex review it, fixes the review comments, and then marks the Codex for review again. We continue like this until Codex approves the code.
This will greatly reduce the number of bugs you get in Claude Code.
The conclusion
In this article, I discussed how to set loops in Claude Code. Loops as a concept have become very fashionable recently because they are a way for coding agents to work autonomously and get more work done. You can set up advanced loops easily by using the /goal command, which causes the agent to continue working toward the goal until it is reached. However, you must also provide the agent with a way to verify their work, which I have covered in this article. I believe that making loops like this is the future of software engineering. You will create agents that work in loops and give them a way to verify their work. This gives you the ability to work on other tasks at the same time. Loops as a concept can also be expanded to have large loops that organize other small loops and self-improving loops as well, which are many terms that I will cover in future articles.
👋 Touch
👉 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
