Skip to content

Cursor: the IDE that finally made AI feel useful

Cursor logo

I've tried a lot of AI coding tools over the last couple of years. Most of them felt like a chat window bolted onto an editor — helpful for a one-off question, but awkward to live in all day. Cursor is the first one I kept.

What it is

Cursor is a code editor built on VS Code. If you've used VS Code before, you'll feel at home immediately: same keyboard shortcuts, same extension ecosystem, same layout. The difference is that AI is woven into the workflow instead of sitting off to the side.

That matters more than it sounds. Context switching kills momentum. When the model can see your open files, your project structure, and the diff you're working on, the suggestions stop being generic and start being actually relevant.

What I use it for

Tab completion is the daily driver. It predicts the next edit — not just the next line — and often nails multi-line changes before I've fully thought them through. It's fast enough that I leave it on all the time.

Inline edits (Ctrl+K) are great for localized changes: rename something, refactor a function, add error handling. You highlight the code, describe what you want, and it edits in place.

Agent mode is where things get interesting. I describe a task — "add a redirect for this URL," "fix the CI workflow," "write a blog post about X" — and it searches the codebase, makes the changes, and runs commands to verify them. It's not perfect, but it's good enough that I reach for it constantly for chores I'd otherwise procrastinate on.

Chat with codebase context handles the exploratory stuff: "how does this plugin work?", "where is the build configured?", "what would break if I changed this?" Having the repo in context beats pasting snippets into a browser tab.

Why not just use ChatGPT in a browser?

You can, and I still do for non-code questions. But for software work, the editor integration wins:

  • The model sees your actual project, not a truncated paste
  • Changes land as diffs you can review before accepting
  • Terminal output, linter errors, and test failures feed back into the loop
  • You stay in flow instead of alt-tabbing between tools

A few tips if you're trying it

  • Start with tab completion and inline edits before jumping into agent mode — they're lower risk and build trust quickly
  • Be specific in prompts: mention files, constraints, and what "done" looks like
  • Review diffs carefully; the model is confident even when it's wrong
  • .cursor/rules (or project rules) help steer behavior for repos you work in often

Worth a look

If you've been curious about AI-assisted development but bounced off tools that felt gimmicky, Cursor is worth trying seriously. It's become my default editor for personal projects, site maintenance, and random scripts.

If you want to check it out, here's my referral link: cursor.com — try Cursor

(Full disclosure: that's a referral link. If you sign up through it, I may get a small credit. I'd recommend it either way.)