, coding agents have grow to be an increasing number of prevalent. Initially, coding agents could only auto-complete specific lines of code. We then experienced how agents could interact with a single file and make changes to entire functions. After this, we began seeing agents able to keeping track of and updating code in multiple files.
Now, coding agents are extremely capable and might work across multiple code repositories, even implementing entire features without having for human intervention.
The capabilities of coding agents have opened up an entire recent world of productivity for software engineers. In this text, I’ll highlight how coding agents have increased my productivity as an engineer, and the way I leverage coding agents maximally by running multiple in parallel.
I aim to create a high-level overview of what coding agents can do for you and the techniques I utilize to get probably the most out of my coding agents by running them in parallel.
Why run coding agents in parallel
Only a 12 months ago, it was almost unthinkable that you may be programming on multiple projects at the identical time. Programming was generally known as a really high cognitive effort activity, where you had to reduce context switching.
If you wish to take full advantage of coding agents, you must run them in paralell. And if you happen to’re not taking full advantage of coding agents, you’re falling behind
I still recommend minimizing context switching. Nevertheless, the capabilities of coding agents have gotten thus far that if you happen to don’t run multiple in parallel, you’re falling behind.
When spinning up a coding agent, you often start it on a task by giving it some directions and asking a couple of questions. After this, nevertheless, the agents start working, and it may take 5-20 minutes before you must interact with the agent again. As an alternative of waiting for this long, you spin up one other coding agent. You may then proceed this cycle of spinning up recent agents until you may have to interact with the primary agent again.
Simply put, the explanation you need to run multiple agents in parallel is that that is the strategy to achieve maximum effectiveness as a software engineer. You might, for instance, have a look at the creator of Claude Code, Boris Cherny, on X.
He posted a thread about how he utilizes Claude Code, where he highlights how he runs 10-20 agents in parallel at any given time.
My parallel coding framework
On this section, I’ll highlight my framework for working with multiple coding agents in parallel. It’s a straightforward four-step process
- Find probably the most high-impact task you may do
- Open up Claude in plan mode, and confer with Claude how you may solve this task. Be happy to have an extended discussion here, and spend quarter-hour on properly shaping the duty. Be certain that the agent has all of the permissions it needs, so it doesn’t interrupt you
- Find the second-highest impact task, start planning, and spin up this agent
- Proceed down your list of high-impact tasks. When you may have to interact with the primary agent again, you may stop spinning up recent agents and check out to finish the tasks you’re working on.
There are three fundamental requirements if you wish to use this framework.
- You would like a very good list of prioritized tasks, based on a value-effort graph
- You ought to be using plan mode liberally
- It is best to have a straightforward way of spinning up many agents in parallel. I exploit Claude Code, though there are various other options on the market
I’ll now have a bit highlighting each of those requirements.
Keeping an inventory of prioritized tasks
This requirement is, for my part, severely underrated. It is best to, at any given point, keep an inventory of probably the most high-impact tasks you may do. Preferably, you furthermore may mix this with the trouble required to finish the duty, and you may have a straightforward priority list.
I feel keeping such an inventory has at all times been vital. It’s easy to finish quite a lot of tasks nowadays, simply using tools like Claude Code. Nevertheless, if you happen to’re not working on the highest-impact tasks, it doesn’t really matter. I often consider the quote below from Elon Musk when prioritizing tasks.
Essentially the most common mistake of a sensible engineer is to optimize a thing that shouldn’t exist
Thus, you have to be using project management tools corresponding to Todoist, Notion, Monday, or similar to arrange and type tasks based on the impact they’ve. Having this list will then make it much easier when you wish to spin up a brand new agent, as you just pick a very powerful task out of your prioritized list.
Use plan mode liberally
A typical mistake I see people make when interacting with coding agents is to provide it vague instructions and easily spin it up. This simply put doesn’t work.
You must provide your agent with elaborate and detailed instructions that, with minimum ambiguity, tell the agent what to do.
The very best strategy to do that is to make use of plan mode and never be afraid to spend time in plan mode. Each time I start a brand new agent on an advanced task. I’m not afraid to spend 20 minutes simply chatting forwards and backwards with my agent on methods to do an implementation and methods to handle edge cases.
The 20 minutes you spend interacting along with your agent are easily regained, since your agent will make fewer mistakes. Moreover, you’ll must iterate less along with your agent after the implementation is completed, to have the precise solution you desire. Plan mode saves you time in the long term.
The identical concept applies to the LLM you’re using to code with. Some persons are tempted to make use of cheaper and faster models when implementing some features.
I’d argue, nevertheless, that this simply finally ends up costing more and taking more time than if you happen to used a bigger and dearer model.
It is because, though the larger model will take longer to provide you with an initial solution, the agent will make fewer mistakes, and also you’ll spend less time having to iterate with the agent to get the precise solution you wish.
Tool to spin up multiple agents
The last requirement I highlighted was to have a straightforward tool to spin up multiple agents. Up until recently, I used to be using Cursor to run my agents. I spotted , nevertheless, that Cursor isn’t probably the most optimal tool relating to spinning up multiple agents, because it’s hard to have a straightforward overview of all of your agents. (Cursor still has its advantages in other situations, nevertheless, so I’m not counting it out).
For running parallel agents, I thus recommend using a CLI-based tool like Gemini CLI or Claude Code. Moreover, I exploit Warp terminal. With this setup, I can have one terminal window open and have all my coding agents running from this terminal.
I can then split my terminal into multiple tabs and spin up my agents. You may see what my terminal looks like within the image below:
Conclusion
In this text, I’ve covered methods to run multiple coding agents in parallel. I discussed why you must run multiple agents in parallel, highlighting that that is how you may achieve maximum efficiency as an engineer. Moreover, I covered specific techniques I utilize to be effective with multiple agents. The fundamental points are to maintain a very good list of prioritized tasks, use plan mode liberally, and have a CLI tool to spin up agents with. I feel coding agents are the longer term of coding, and spinning up multiple agents in parallel needs to be the default behaviour for quite a lot of programmers on the market. Should you haven’t began already, you need to begin practicing coding in this fashion.
👉 My Free Resources
🚀 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:
🧑💻 Get in contact
✍️ Medium
