Cursor is doing something VS Code still cannot

I have been writing automation scripts and configuration files in homelabs long enough to know when a tool removes friction instead of just claiming to. Cursor does that. Not because it’s faster or shinier. Because it understands your whole project at once, and it does not make you ask permission to use that knowledge.

The difference between completion and awareness

Every editor has autocomplete now. GitHub Copilot works fine. But there is a gap between “the AI saw the last thirty lines of your file” and “the AI read your entire codebase and knows why you structured it that way.”

I noticed this building a Home Assistant automation in YAML. Standard setup: services, automations, scripts split across three files. I asked Cursor’s chat to add a new automation that follows the existing naming convention and hooks into the current service pattern. It did not ask me to paste the files. It opened them. Read the structure. Saw that I use a delayed_notification helper in three places and applied the same pattern to the new automation. Then it edited all three files at once.

That is not Copilot. That is not VS Code. That is different.

Where it actually matters

Docker Compose files. Ansible playbooks. Helm charts. These are things homelabbers write constantly and they benefit hard from understanding scope.

I had a Compose file with seven services, a custom network, three volumes. I asked Cursor to add Traefik as a reverse proxy and wire all the internal services to it. Instead of breaking it or asking me to do half the work, it:

• Added Traefik service with sensible defaults
• Updated each existing service with labels
• Understood which services should be exposed and which should stay internal
• Did not break the existing network

The file was coherent. It worked the first time. That is not luck.

What actually stopped me cold

It is not self-hosted. It is not open source. It requires an API key and it phones home. If you run your entire homelab offline, this does not work for you. Full stop.

I spent a week assuming I would hate that constraint. I do not love it. But I also had to admit: running it locally would lose the thing that makes it work. The model is trained on enough public code to understand patterns. Local LLMs are not there yet. Not by a wide margin.

That is frustrating. I own the choice anyway. Some tools are good enough that the tradeoff makes sense.

The honest part

Cursor is not magic. It hallucinates. It writes bad YAML sometimes. It does not read your mind about what your network setup actually supports. You still need to think. You still need to test. You still need to know enough to catch when it is wrong.

What it does is stop you from holding entire files in your head while you type. It stops you from manually copying patterns across five files and missing one. It stops you from writing the boilerplate that you have written seventeen times before.

For the specific work of building and maintaining homelab infrastructure, that is enough. I have tried to go back to raw VS Code three times since I started using Cursor. Each time I lasted about an hour before I got tired of not having the context.

The question is not whether Cursor is good. It is whether losing the offline-only guarantee is worth the weeks of your life it buys back. For me, it is. Your call will depend on what your homelab is actually for.

Explore Cursor in our AI Homelab Toolkit.