different coding setups people use for programming. In this text, I’ll take you thru my personal coding setup and the tools and applications that I exploit to attain maximum efficiency when programming.
It is a setup that I’ve created through extensive testing and experimenting myself through trial and error. While testing, I’ve attempted to make use of several different applications for programming, and every of them has benefits in several settings.
I’ll take you thru the present coding setup I even have, though it’s, after all, subject to alter soon in the longer term with the rapid advancement of LLM technology.
I’m not sponsored by any of the tooling mentioned on this video, and it’s simply the tooling I exploit on a day-to-day basis as a programmer.
Why optimize your coding setup
As a programmer, your coding setup is one of the crucial essential elements you’ll be able to optimize. That is where you spend most of your time solving different problems. Due to on a regular basis you spend together with your coding setup, you need to spend time ensuring it’s optimized in your personal workflows.
Personally, I at all times search for opportunities to make my setup more efficient. For an extended time frame, I used Cursor every day because the platform from which I did all my coding. A number of weeks ago, I suddenly shifted to using purely Claude Code through Warp, which essentially makes up nearly all of my coding setup.
The switch from using Cursor to using Claude Code through Warp was one of the crucial significant productivity increases I’ve experienced since I first noticed how efficient agents could program for me. Warp + Claude Code has helped me massively in my each day work as an information scientist in a startup working on document AI.
Walking through my coding setup
On this section, I’ll walk you thru the various tooling, techniques, and approaches I exploit to optimize my coding setup. I’ll cover the applications I exploit on a day-to-day basis, but additionally how I utilize and get probably the most out of those applications and other essential tricks I exploit to make my coding as effective as possible.
All of the ideas I’ll cover on this section have a big impact on my productivity as an engineer.
Tooling
To start with, I need to cover the tooling I exploit. I exploit Claude Code and Warp for just about all of my coding. If I need to envision some production logs or if I need to repair a bug or implement a brand new feature, I’ll essentially at all times use Claude Code in Warp.
Inside Warp, I even have the next setup. I even have each tab in Warp as a separate folder I’m working in. So if I’m working in folder A, that’s my first tab in Warp. And if I’m working in folder B, that’s my second tab. Now, I typically find myself having several agents running inside the same folder. On this case, I make a split pane using CMD + D in Warp, so my tab is split into multiple panes. Depending on the duty I’m working on, I could have as much as five agents working inside the same repository. After which I even have different repositories in several Warp tabs.
I need to notice one exception where I exploit Cursor as an alternative of Claude Code: When I want full control of the code. For instance, if the feature is of critical importance or an element of critical infrastructure. Also, typically once I run essential migration scripts or backfilling scripts, I’ll also do it in Cursor because this provides me more control of the code. I may run the code myself through interactive windows with Python.
Git worktrees
As I discussed in my previous section, I often find myself running multiple agents inside the same repository. If you’ve multiple agents updating files at the identical time in the identical repository, you’ll run into problems with agents colliding with one another. To resolve this problem, you need to use Git Worktrees.
Git worktrees are essentially copies of Git repositories that you could make to have agents run completely separate from one another. So each time I spin up a brand new agent, I tell it to begin a brand new git worktree for what it’s working on. and that agent can now work completely individually from all other agents working in the identical repository.
That is a vital feature if you ought to work with parallel agents in Claude Code (which is considered one of the main advantages of working with Claude Code). Thus, you need to definitely be utilizing Git Worktrees in your day-to-day programming with parallel agents.
Slash commands
Slash commands are one other very powerful feature. Slah commands are essentially stored prompts, so you’ll be able to quickly access a prompt that you’ve stored on a previous occasion. For instance, if you’ve a really repetitive prompt, you need to store it as a slash command. Some examples of this are:
Slash commands are incredibly powerful, and I’ve covered them in considered one of my previous articles. The good thing about slash commands is twofold. To start with, you save time by not having to put in writing out the prompt each time. So as an alternative of getting to put in writing out an extended prompt, telling the model that it must:
- Pull the most recent dev branch and rebase on top of it
- Run precommit checks
- A very good PR description
- Make a pull request from a feature branch to dev
As a substitute of getting to put in writing out all of this, you’ll be able to simply store this prompt in a slash command and access the prompt immediately.
The second advantage is that you simply get to be more consistent when writing your prompts. For instance, when creating pull requests to dev, as I discussed, you would need to run a series of checks (pull latest dev branch, rebase, run precommit checks, …). When you write this out each time, you risk forgetting parts of the prompt. This is just not an issue if you happen to use slash commands, nevertheless, since you’ll at all times be utilizing the identical prompt, and also you’ll be more consistent.
Slash commands make you each faster and more consistent
Low threshold to fireplace off agents
One other topic I need to cover is that you need to have a brilliant low threshold to fireplace off agents to perform tasks for you. At any time when you’ll be able to consider a brand new task or get a brand new problem you’ve to resolve, you need to just fire off an agent. For instance, if I notice a button that’s misaligned, some text in my application that needs to be updated, or translations that need to be updated. I simply fire off a brand new agent, let it run fully autonomously, and create a pull request for me.
The predominant point is that you need to have a low threshold to fireplace off agents since it’s so low cost to run and costs you so little time. The fee of firing a brand new agent is basically spending time writing out a great prompt and, in lots of cases, answering a number of questions the agents have so that you can properly understand the duty you gave them.
There at the moment are many tools on the market that provide quite a lot of token usage for a comparatively low price. For instance, I’m using the $200 Claude Code subscription, which is a set amount per 30 days, and I’ve never run into rate limits. This implies I can fire off as many agents as I can without additional cost.
Utilize the perfect models
One other tip I even have might sound very obvious, but I at all times recommend using the perfect models each time you’re employed with programming. The rationale for that is that in the long run, this protects you each money and time.
Yes, the perfect models are typically the costliest models per token and are also the slowest models. Nevertheless, it seems that if you happen to use cheaper models, they are going to more often make mistakes, which takes additional time so that you can fix and iterate on, which again makes the model utilize much more tokens. Thus, ultimately, it often seems that using a less expensive, smaller model actually seems to be costlier and time-consuming.
It’s best to due to this fact be utilizing the frontier models resembling Gemini 3 Pro, Claude 4.5 Opus, and GPT 5.2 Codex. There are also some up-and-coming open source models performing well on the coding benchmarks, though I haven’t achieved the identical success with open source models as I’ve achieved with frontier closed source models.
Conclusion
In this text, I’ve covered the way to have a maximum efficiency coding setup. I’ve discussed the coding setup I exploit on a day-to-day basis, where I exploit the Warp terminal with Claude Code. Moreover, I exploit specific techniques resembling organizing Warp with split panes and tabs by the folder I’m working on. I’m also ensuring to at all times use the most recent and best coding models. I imagine spending time optimizing your coding setup is a superb use of time. As a programmer, your coding setup is considered one of the belongings you spend probably the most time with, and if you happen to could make that a number of percentage points more efficient, it should likely repay in the long term.
👉 My Free Resources
💌 Substack
🚀 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
