Despite Cursor and Antigravity making the rounds, VS Code is still the only code editor I need for my programming tasks. Sure, it doesn’t have the same bells-and-whistles as its AI-heavy offshoots, but it’s the perfect companion for my local-only setup. In fact, I’ve been using it ever since I got into coding my own apps, and despite some minor quirks, I doubt I’ll be ditching VS Code anytime soon.

That said, my old VS Code had started to feel somewhat sluggish as of late, and the extra visual elements I’d added over the years didn’t make things any better, either. So, I nuked my VS Code instance and rebuilt it from scratch to optimize my coding setup – and it turned out to be the best productivity decision ever.

My VS Code instance had far too many extensions

I didn't need half of them for my coding tasks

Considering all the home lab containers and VMs I deploy on a whim, it shouldn’t come as a surprise that I armed VS Code with more extensions than I could ever need. On the language front, I’d configured multiple web development frameworks, data manipulation query languages, and Infrastructure-as-Code extensions. But I’d only use a specific language (or maybe two) for my coding workloads, while leaving the rest of the extensions gathering virtual dust on my VS Code instance.

However, the real extension bloat came from the visual elements and AI-centric tools I’d installed during random bouts of tinkering frenzy. I had far too many extensions for altering VS Code’s appearance, while all I really needed was a simple formatter like Prettier to make my weirdly-indented code blocks more bearable to read.

The same applies to the seemingly QoL-enhancing extensions I’d configured on VS Code. Don’t get me wrong, I still consider Container Tools, Remote-SSH, Dendron, and Live Server borderline essential for VS Code. But since extensions like Trailing Spaces and Indent Rainbow can easily be replicated using simple VS Code settings, they just made the code editor more bloated without bringing anything useful to the table. And the situation was largely the same with my AI-centric extensions.

Integrating local LLMs with VS Code increased my productivity multifold

And I’ve also chucked the built-in Copilot “features” out of my code editor

Although GitHub Copilot and its cloud-based rivals are great for vibe-coding, I tend to stick to my local LLMs for my VS Code misadventures. Between their privacy-respecting nature and lack of subscription fees, self-hosted models are perfect for my coding tasks. Plus, the likes of Gemma4-26B-A4B and Qwen3.6-35B-A3B can hold their own against typical cloud-based models, and they mesh incredibly well with VS Code for everything from troubleshooting terminal logs and scanning code for vulnerabilities to applying indentation modifications to config files and debugging faulty snippets. As such, the Copilot tools that ship with VS Code are borderline useless for me, and all they do is add extra options in the context menu. So, I removed them altogether on my fresh VS Code instance.

Unfortunately, most of the AI-centric extensions I’ve encountered on the marketplace were primarily focused on offering proprietary models (and charging subscriptions for API usage), and only featured local LLMs as an afterthought. Continue was the only AI extension I’d use on VS Code for months, as it meshed well with my local LLM, but I’ve since moved on to llama-vscode. Credit where it’s due, Continue was brilliant during my days as a fledgling Ollama user, but I started relying on llama-vscode once I switched to llama.cpp as the inference engine. Sure, it can take a while to get accustomed to llama-vscode’s somewhat barebones interface, but it’s far more powerful than it initially appears.

Since I’ve got different LLM pipelines configured on my home lab nodes, I’ve paired llama-vscode with separate models for autocompletion, embedding, and chat tasks. The llama-vscode extension plays just as nicely with the dozens of tools on the MCP servers for my FOSS utilities.

Turns out, less is more when it comes to VS Code extensions

As much as I love tinkering with different VS Code extensions, it was about time I got rid of the excess bloat plaguing my code editor. Besides a couple of QoL-enhancing extensions and the coding languages I typically work with, Prettier and llama-vscode are all I need for an efficient VS Code setup – one that consumes fewer system resources and doesn’t distract me with useless UI elements.