Cursor
The AI IDE that made me stop pretending I didn’t need help.
Why I'm Starting Here
I've been writing code since before "vibe coding" was a phrase anyone was allowed to say in public. I shipped enterprise .NET back when serializing an object across a wire was a religious debate, ground out an MCSD when that meant something, built voxel terrain in Unity for fun, and somewhere in the middle of all that I started a hardware company called PowerBobber that ended up on a stage with Intel because we figured out how to make a fishing lure smarter than most apps.
So when I tell you Cursor changed how I work, understand the bar that's clearing. I am not a guy who needs an editor to hold my hand. I needed an editor that could keep up.
Cursor keeps up.
It's VS Code with a brain bolted onto it. If you've ever used VS Code, you already know the muscle memory — same shortcuts, same extensions, same panel layout, same everything. The difference is that the AI sitting inside it has actually read your codebase. Not "trained on code in general." Read yours. Asks intelligent questions. Edits across files. Catches things I miss at 1 AM.
That's the whole pitch. The rest is detail.
Who Should Actually Use This
You, if you already write code. Stop reading this paragraph and just install it. You're going to be 2-3x faster within a week. I'm not exaggerating and I'm not selling you anything — I pay for it like everybody else.
You, if you're a technical founder building a product. This is the tool. Full stop. Frontend, backend, database, deployment scripts — Cursor handles the context-switching better than your brain does at hour ten of a build session.
You, if you're a hustler with no CS degree but a real project and real patience. Cursor will not coddle you. You'll see error messages. You'll have to learn what npm install means. But the ceiling is the sky, and you own every line. If you learned Excel by breaking spreadsheets until they worked, this is the same energy.
You, if you want a chatbot that builds your app while you sip coffee. Wrong chapter. Skip ahead to Lovable or Bolt. No shame — different tool for a different job.
What Actually Matters
I'm cutting the marketing list. Here's what I use every day.
Cmd+K is the magic. Highlight code, hit Cmd+K, type what you want changed in plain English, see a diff, accept or reject. That tight loop — describe, preview, ship — is the entire reason I switched. It's faster than typing the change myself, and it's correct more often than I am at midnight.
Cmd+L is the room you talk to. Sidebar chat with full codebase awareness. @filename to point at a specific file. @codebase to make it think across the whole project. @docs to pull in library docs. @web to grab live information. When I'm debugging a weird auth issue, I describe the symptom and Cursor finds the file before I'd have finished opening the folder.
Composer + Agent mode. This is where it stops being autocomplete and starts being a junior engineer. "Add Stripe checkout, use the existing user model, write the webhook handler" — Composer plans the change across files, Agent runs the terminal commands, reads errors, fixes them, and reports back. I still review every diff. But the loop of code-test-fix-repeat? Compressed.
.cursorrules. I will die on this hill. More on it below.
Tab completion that reads the room. It learns your project. By week two it's predicting the next three lines correctly because it noticed how you named your variables in the file you wrote yesterday. Spooky and useful.
Pick your model. Claude, GPT, Gemini, Grok, or Cursor's own Composer 2 — their in-house model, and since the March 2026 update it's genuinely frontier-fast. Different models have different vibes. I use Claude for reasoning, Composer 2 for boilerplate at speed. Switch mid-session. No drama.
What It Costs
Free tier exists. It's a trial. Don't kid yourself.
Pro is $20/month and that's the answer for 95% of you reading this. It's a $20 credit pool now — Auto mode is unlimited, credits only burn when you hand-pick a frontier model. Two coffees. Get over it.
Pro+ is $60 and Ultra is $200. Same deal, 3x and 20x the usage. If you're running agents all day every day, Ultra pays for itself. If you have to ask, you don't need it yet.
Teams is $40/user/month with privacy mode (your code isn't training anyone's model), Bugbot code reviews, and admin controls. If you're building anything you'd be upset to see in a competitor's product, this tier or higher.
Enterprise is the procurement-paperwork tier. If you need it, you know.
Heads up: Agent mode burns credits fast when it's looping on a hard problem. Watch the meter during sprints. I've hit the wall at the worst possible moment more than once. Pace yourself or upgrade.
What It's Good At
Web apps: Best in class. Next.js, React, Node, Python APIs, full-stack — this is Cursor's home. Five stars and I'm not handing those out.
Mobile: Fine for React Native. Fine-ish for native Swift/Kotlin — works, but you can feel the AI is less specialized in those ecosystems. If mobile is your primary target and you want native, look at dedicated tooling.
Scripts, CLIs, automations, desktop utilities: Excellent. I write more throwaway Python in Cursor than I used to write in a year. Half of them turned out to not be throwaway.
Hosting? That's Your Problem
Cursor is a local editor. It doesn't host anything. Your code lives on your machine until you push it somewhere.
This is a feature, not a bug. You're not married to Vercel or Supabase or anybody. Pick whatever you want. I usually run Vercel for the frontend, Railway for backends and Postgres, Supabase when I want auth handed to me on a plate. None of that is Cursor's job, and Cursor will happily help you write the deploy scripts for any of it.
If you're new and that paragraph stressed you out: Replit handles all this for you (Chapter 9). Worth knowing your options.
The Story
Real one. I was building the prototype for one of our internal tools at AIEA — the kind of thing that would normally take two weeks of part-time effort. I had a dental appointment in the morning and a flight that afternoon. I sat down at 7 AM, opened Cursor, and started describing what I wanted: data ingestion, a small dashboard, a way to flag anomalies, deploy it somewhere I could share a link.
I was not "vibe coding." I was directing. Cursor wrote, I corrected, Cursor adjusted, I deployed. By the time I left for the dentist, the thing was live. By the time I landed, two people on the team were already using it and finding bugs (which I then fixed on the plane).
That used to be a sprint. Now it's a morning.
The point isn't that I'm fast. The point is that the floor moved. What used to be hard is now ordinary. Which means the bar for what's worth attempting just went up — and that's the actual unlock.
Where It Bites
It's confidently wrong sometimes. It will hand you a function that looks great and quietly assumes a library version that doesn't exist anymore. Read the diffs. Don't accept changes you didn't read. This is the single most expensive lesson new users learn, and I see it weekly when I'm reviewing other people's code at CodeDR.AI.
It will get stuck in loops in Agent mode. Watch the token meter. If it's "thinking" for three minutes, kill it, reword the prompt, start over. The model is not going to figure it out by trying harder — you are.
And as I said, hosting is on you. That's the deal.
One more thing: SpaceX bought Cursor's parent company this summer. Sixty billion dollars, all stock. Yes, that SpaceX. The editor hasn't changed and the Grok limits got noticeably generous, but when your IDE is owned by a rocket company, you keep one eye on the roadmap. I'm not switching. I'm just watching.
Gavin's Rule: .cursorrules On Day One
Before you write a single line of real code, drop a
.cursorrulesfile at the root of your project. Tell the AI what stack you're using, what conventions you follow, what libraries to prefer, what to never do. Ten lines. Fifteen. Doesn't need to be a novel."TypeScript strict mode. Tailwind for styling. Never use class components. We use Drizzle for the ORM. Don't suggest jQuery, don't suggest moment.js, don't suggest any library older than my last divorce."
Every interaction in that project respects those rules forever. This is the difference between an AI that feels onboarded and an AI that keeps trying to install Bootstrap. Skip this and you'll spend your first week undoing assumptions. Do it and the AI feels like it actually works for you.
If you take one thing from this chapter, take this. — and yeah, this is exactly the kind of thing CodeDR.AI catches in code review, but you can also just do it yourself for free in five minutes. Do it.
Bottom Line
If you write code, this is your editor. If you're learning to write code and you've got patience, this is the one with the highest ceiling in this whole guide. You'll work harder than you would in Lovable. You'll also build things Lovable can't touch.
— Gavin