Right now everyone is chasing the perfect AI-powered coding setup. They've got a dozen different extensions, a custom-trained model that hallucinates in iambic pentameter, and a workflow so complex it requires its own README. But let's be honest, you don't need all that.
The beauty of these AI agents is that they reduce the complexity of our workflows, yet people love adding complexity to stuff. As an experienced dev, instead of writing the massive amount of boilerplate needed to prototype this Next.js app and build a CI, I can just describe what I need, then review and approve.
All you really need are two things: Cursor and Gemini CLI.
Cursor is just a fork of VS Code that's been elegantly rigged for AI. It gets the inline-diffing and code generation right. The Gemini CLI is a terminal agent that can see your files and talk to your dev server. Between the two, you have 80% of what you could possibly want from an AI assistant. One editor for your code, the other runs it, analyzes the results, and does some filesystem operations.
Most devs overthink and over-optimize. The cost of messing with tooling and toolchains is so high that you end up benchmarking LLMs when you could be building.
Get back to actually building something.
Cursor
Cursor is a fork of VS Code that's been elegantly rigged for AI. It gets the inline-diffing and code generation right. This lets you fine-tune and iterate when the CLI is stuck in a loop or adding weird dependencies or something.
Gemini CLI
The Gemini CLI is a terminal agent that can see your files and talk to your dev server. You can have it analyze an entire project and act as a junior dev, writing you a feature proposal or plan.