My experience with Vibe Coding
Think of being a technical project manager with a dedicated team member who has solid coding skills. While this AI team member is well trained in code syntax and some best practices, it still behaves like someone who doesn’t see the whole picture. It focuses only on the task at hand. To me, that is the current vibe coding experience.
This experience will vary for coders depending on their background and the tools they use. I started testing different models within the ChatGPT app, keeping it open while I worked on personal projects in VS Code. ChatGPT is impressive, especially in how it reads the context from your open editor to suggest relevant code. That integration is useful, though it still requires switching between apps.
VS Code has Copilot, which now supports multiple large language models like OpenAI’s GPT-4, Google’s Gemini 1.5 Pro, and Anthropic’s Claude 3.5 Sonnet. I use VS Code often because it’s been around for a while, and it’s what I use at work alongside Visual Studio. That said, I’ve never completely liked how Copilot interacts with the editor. It has improved, and with new support for image inputs and expanded model choices, it’s worth revisiting. For example, you can now upload screenshots to Copilot Chat, and the model can interpret them to offer coding assistance. That’s a big step forward in usability.
On my personal Mac, I use JetBrains Rider, a full IDE that’s great for large and complex projects. Like other modern editors, Rider supports chat-based interactions with capable LLMs. It’s my go-to for heavy-lifting on macOS.
Both VS Code and Rider are excellent, but I have to admit that the new kid on the block handles AI chat integration better than anything else I’ve used. I’m talking about Cursor AI. I first found it months ago, gave it a quick try, and didn’t understand the hype. I stopped using it until last week, when I heard about new model announcements and saw Cursor mentioned again online. I opened it back up for a new personal project and gave it a proper try.
After a few days of using Cursor, while also working on the same code in VS Code and Rider, it was clear that Cursor offers a smoother, more useful integration with LLMs. It explains code changes well, and while the models still make mistakes, even bad ones sometimes, the way the editor handles those changes is great. You can preview edits directly, and the overall experience is fast and lightweight. It reminds me of what it felt like to use VS Code for the first time.
Cursor recently added features like Composer history, letting you track and edit past sessions more easily. That’s a nice touch when you’re iterating quickly on code. Reports even suggest developers see a 30 percent speed boost for routine tasks when using Cursor. That definitely aligns with my experience so far.
My second serious go at vibe coding has been a lot more rewarding. Cursor, combined with Claude and occasionally GPT models, has created a great workflow that fits how I like to work.
The project I’m working on right now is an article reader. If you’re a Pocket user, you probably heard that Mozilla is shutting it down. That’s unfortunate. Pocket has been one of my most-used apps, great for reading articles in a clean, focused layout without ads or distractions. With it going away, I decided to build my own version. It’s a simple web app that lets you paste a URL, read the article in a minimalist format, and save it for later. I’ll share more about it soon.
For now, back to vibe coding and making the most of these tools.
Have a great weekend!
