Cursor is brimming with features and models right now. Figuring out how to actually use it well isn't obvious.

I'm a somewhat experienced developer taking time off work, and I've been using Cursor Ultra 8+ hours a day building out my personal project. Here's what I've found:

Composer is king for edits

Fast, accurate, cheap. But it's not great at decision making, so you need other models to set up context and plan properly before letting Composer do the actual implementation.

Plan mode vs. Ask mode

Plan mode is nice in theory but buggy and inconsistent. It's also bad for interrogating a plan. You lose the conversational back-and-forth that LLMs are good at, especially when refining a feature.

Ask mode is where it's at. Any time I want to make a complex change, I start there with a smarter model. Gemini Pro 3 is great when it works, but it often doesn't for complex tasks (loop errors, capacity issues, throttling). GPT-5.2 has been my go-to. Good speed, quality, and availability.

My workflow

I start in ask mode, discussing the feature: which files it'll touch, what I want, how I think I want it. Then I add this to the end:

With all this in mind, can you evaluate what needs to be done, identify the changes, and whether anything needs to be reworked to ensure a clean design?

That last part matters for my project where I can afford constant re-architecting. It discourages models from taking shortcuts or building on top of bad decisions made elsewhere.

GPT-5.1 will come back with code examples, explanations, and rationale. You can dig into the proposed solution way more than you could with a high-level .md from Plan mode, and steer it where you need it to go. By the time you switch to edits, there's no ambiguity about what needs to happen.

Once I'm happy with the plan, I switch to Agent mode with Composer and just say "Let's implement." 8 times out of 10 it's exactly what I need.