coding agents are becoming increasingly common. Originally, coding agents were able to auto-complete certain lines of code. We then see how agents can interact with a single file and make changes across operations. After this, we started to see agents that could track and update code in multiple files.
Now, coding agents are extremely capable and can work in all code repositories, even implementing all features without the need for human intervention.
The capabilities of coding agents have opened up a new world of productivity for application developers. In this article, I will highlight how coding agents have increased my productivity as a developer, and how I use coding agents more by using parallel iteration.
I aim to create a high-level overview of what coding agents can do for you and the strategies I use to get the most out of my coding agents by using them in tandem.
Why use encoding agents in parallel
Just a year ago, it was almost unthinkable that you could be planning multiple projects at once. Editing was known to be a very high cognitive effort task, where you had to minimize content changes.
If you want to take full advantage of coding agents, you need to use them in tandem. And if you're not making full use of coding agents, you're falling behind
I still recommend that you limit the content changes. However, the power of coding agents has come so far that if you're not running a lot in parallel, you're falling behind.
When hiring a coding agent, you usually start them off by giving them some directions and asking a few questions. After this, however, the agents start working, and it may take 5-20 minutes before you need to contact the agent again. Instead of waiting that long, you spin another coding agent. You can then continue this cycle of spinning new agents until you have to contact the first agent again.
Simply put, the reason you should use multiple agents in parallel is that this is the way to achieve maximum success as a software developer. For example, you can look at the creator of the Claude Code, Boris Cherny, on X.
He posted a thread about how he uses the Claude Code, where he highlights how he runs 10-20 agents in parallel at any given time.
My corresponding code framework
In this section, I will highlight my framework for working with multiple agents for parallel coding. It's a simple four-step process
- Find the most productive work you can do
- Open Claude in program mode, and talk to Claude about how to solve this task. Feel free to have a long chat here, and spend 15 minutes on training the job properly. Make sure the agent has all the permissions it needs, so it doesn't interfere with you
- Find a second high impact job, start planning, and investigate this agent
- Continue down your list of high-impact activities. If you have to contact the original agent again, you can stop spinning new agents and try to complete the tasks you are working on.
There are three important requirements if you want to use this framework.
- You need a good list of important tasks, based on an effort-value graph
- You should use program mode freely
- You should have an easy way to spin multiple agents in parallel. I use Claude Code, although there are many other options out there
I will now have a section highlighting each of these requirements.
Keeping a list of important tasks
This need, in my opinion, is greatly underestimated. You should, at any given time, keep a list of the most impactful activities you can do. Better yet, you also combine this with the effort required to complete the task, and you have a simple priority list.
I think keeping such a list has always been important. It is easy to complete many tasks these days, using tools like Claude Code. However, if you don't work in high-impact jobs, it doesn't really matter. I often think of the quote below from Elon Musk when I prioritize tasks.
The most common mistake of a smart developer is to make something that shouldn't exist
Therefore, you should use project management tools like Todoist, Notion, Monday, or similar to plan and schedule tasks based on their impact. Having this list will make it much easier if you want to investigate a new agent, as you simply select the most important job from your prioritized list.
Use program mode freely
A common mistake I see people make when contacting coding agents is to give vague instructions and just screw it up. This simply does not work.
You need to give your agent detailed and detailed instructions on how to, with minimal ambiguity, tell the agent what to do.
The best way to do this is to use editing mode and don't be afraid to spend time in editing mode. Whenever I start a new agent in a difficult task. I'm not afraid to spend 20 minutes just talking back and forth with my agent about how to get started and how to handle edge cases.
The 20 minutes you spend working with your agent will be well spent, as your agent will make fewer mistakes. In addition, you will have to iterate a bit with your agent after the implementation is complete, to have the exact solution you want. Editing mode saves you time later.
The same concept applies to the LLM you use to code it. Some people are tempted to use cheap and fast models when doing other things.
However, I would argue that this ends up being more expensive and time consuming than when using a larger and more expensive model.
This is because, although a large model will take longer to come up with the first solution, the agent will make fewer mistakes, and you will spend less time iterating with the agent to find the exact solution you want.
Multi-agent integration tool
The last requirement I highlighted was to have a simple tool to integrate multiple agents. Until recently, I was using Cursor to run my agents. I have noticed, however, that Cursor is not the most suitable tool when it comes to spinning multiple agents, as it is difficult to have a complete view of all your agents. (The cursor still has advantages in some cases, though, so I'm not counting it).
For using similar agents, I therefore recommend using a CLI-based tool such as Gemini CLI or Claude Code. In addition, I use the Warp terminal. With this setup, I can open a single terminal window and have all my coding agents run from this terminal.
I can then split my terminal into multiple tabs and browse my agents. You can see what my terminal looks like in the image below:
The conclusion
In this article, I have covered how to use multiple coding agents in parallel. I discussed why you need to use multiple agents in parallel, highlighting that this is how you can get maximum efficiency as a developer. In addition, I have compiled some strategies that I use to work with multiple agents. The main points are to maintain a good list of prioritized tasks, use program mode freely, and have a CLI tool with which to connect agents. I think coding agents are the future of coding, and spinning up multiple agents in parallel should be the default behavior for most programmers out there. If you haven't started yet, you should start practicing coding this way.
👉 My Free Resources
🚀 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:
📩 Subscribe to my newsletter
🧑💻 Contact
🐦 X / Twitter
✍️ In the middle
