setting up different codes that people use to edit. In this article, I'll take you through my personal coding setup and the tools and apps I use to achieve maximum efficiency when coding.
This is a setup I created through extensive testing and testing myself through trial and error. While testing, I tried using several different editing apps, and each of them has advantages in different settings.
I will walk you through the current coding setup I have, although, of course, it may change in the near future with the rapid development of LLM technology.
I am not sponsored by any of the tools mentioned in this video, and they are tools I use every day as an editor.
Why are you configuring your code setup?
As a programmer, your code setup is one of the most important things you can prepare for. This is where you spend most of your time solving different problems. Because of all the time you spend setting up your code, you should spend time making sure it's optimized for your personal workflow.
Personally, I'm always looking for opportunities to make my setup work better. For a long time, I used Cursor daily as a platform where I did all my coding. A few weeks ago, I suddenly switched to using Claude Code with Warp, which does most of my coding setup.
Switching from using Cursor to using Claude Code using Warp was one of the most significant productivity increases I've experienced since I started realizing how efficient agents can program for me. Warp + Claude Code has helped me a lot in my daily work as a data scientist early on working on document AI.
Walking through my code setup
In this section, I'll walk you through the different tools, techniques, and methods I use to improve my code setup. I'll include the apps I use every day, but also how I use and get the most out of these apps and some key tricks I use to make my coding as efficient as possible.
All the tips I will cover in this section have a huge impact on my productivity as a developer.
Tools
First, I want to cover the tools I use. I use Claude Code and Warp for almost all of my coding. If I want to check some production logs or if I want to debug or implement a new feature, I will actually use Claude's Code in Warp.
Inside Warp, I have the following setup. I have each tab in Warp as a separate folder that I work on. So when I'm working in folder A, that's my first tab in Warp. And if I'm working on folder B, that's my second tab. Now, I often find myself with several agents running within the same folder. In this case, I create a split pane using CMD + D in Warp, so my tab is split into multiple panes. Depending on the job I'm working on, I can have up to five agents working in one location. Then I have different repositories in different Warp tabs.
I want to note something different when I use Cursor instead of Claude Code: If I need full control of the code. For example, if the feature is very important or is part of a critical infrastructure. Also, usually when I use important transport scripts or backfill scripts, I'll also do it in Cursor because this gives me more control over the code. I can also run the code myself using interactive windows with Python.
Git work trees
As I mentioned in my previous section, I often find myself running multiple agents on the same endpoint. If you have multiple agents updating files at the same time in the same environment, you will have problems with agents conflicting with each other. To solve this problem, you can use Git Worktrees.
Git worktrees are essentially Git backups that you can create so agents can run completely independently of each other. So whenever I spin up a new agent, I tell it to start a new git worktree for what it's working on. and that agent can now run completely separately from all other agents running on the same endpoint.
This is an important feature if you want to work with parallel agents in Claude Code (which is one of the biggest advantages of working with Claude Code). Therefore, you must be using Git Worktrees in your daily projects with similar agents.
Slash Commands
Slash commands are another very powerful feature. Slah commands are stored commands, so you can quickly access information you've saved in the past. For example, if you have a repeating message, you should save it as a slash command. Some examples of this are:
Slash commands are incredibly powerful, and I've covered them in one of my previous articles. The benefit of slash commands is twofold. First, you save time by not having to write information all the time. So instead of writing a long prompt, tell the model that it needs:
- Pull the latest dev branch and downgrade onto it
- Do a pre-commitment check
- Good description of PR
- Make a pull request from the feature branch to get started
Instead of typing all this, you can just save this notification in the slash command and access it quickly.
The second benefit is that you get consistency when writing your information. For example, when you create pull requests to launch, as I said, you will have to do a series of tests (pull the latest dev branch, rollback, do previous tests, …). If you write this down all the time, you run the risk of forgetting parts of the information. This isn't a problem if you use slash commands, though, because you'll always be using the same information, and you'll be more consistent.
Slash commands make you both fast and consistent
Lower threshold for switching off agents
Another topic I want to bring up is that you should have a very low threshold for firing agents to do jobs for you. Whenever you can think of a new job or find a new problem to solve, you should just fire the agent. For example, if I see a button that isn't properly indented, some text in my app that needs to be updated, or a translation that needs to be updated. I simply fire up a new agent, let it run automatically, and make a pull request.
The bottom line is that you should have a low threshold for firing agents because they are cheap to use and cost very little time. The cost of firing a new agent is actually spending time writing good information and, in most cases, answering a few questions the agents have so you can better understand the job you've given them.
Now there are many tools out there that offer the use of multiple tokens at a very low price. For example, I use $200 of the Claude code, which is a fixed amount per month, and I have never met the rating limits. This means I can fire as many agents as I can at no extra cost.
Use the best models
Another tip I have may sound very obvious, but I always recommend using the best models whenever you are working with a program. The reason for this is that in the long run, this saves time and money.
Of course, the best models are usually the models that cost the most per token and are also the slowest models. However, it turns out that if you use cheap models, they will make mistakes more often, which takes more time to correct and iterate, which also makes the model use even more tokens. So, in the end, it often turns out that using a cheaper, smaller model actually turns out to be more expensive and time-consuming.
So you should use borderline models like Gemini 3 Pro, Claude 4.5 Opus, and GPT 5.2 Codex. There are also up and coming open source models that work well for coding benchmarks, although I haven't achieved the same success with open source models as I have with borderline closed source models.
The conclusion
In this article, I have covered how to have a more efficient coding setup. I discussed the code setup I use every day, when I use the Warp terminal with Claude Code. In addition, I use some techniques such as organizing Warp with separate panes and tabs for the folder I'm working on. I also make sure that I always use the latest and best coding models. I believe that spending time preparing your code setup is a very good use of time. As a programmer, your code setup is one of the things you spend the most time with, and if you can make those few points work well, it will probably pay off in the long run.
👉 My Free Resources
💌 Stack
🚀 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:
🧑💻 Contact
🐦 X / Twitter
✍️ In the middle
