Cursor AI, tested from the sign-up screen inward

I didn't sit down with Cursor as a believer or a skeptic. I sat down as someone who's a little wary of any tool that promises to write my code for me, and who wanted to see how it behaved before trusting it with anything that mattered.

So I did what I do with every new product: I signed up with a throwaway email I keep for exactly this, and I paid attention to the small stuff first. How does it handle my login? What does it switch on by default? Where does it draw the line between “free” and “give us your card”? Those little moments usually tell you more about a company than any landing page does.

This is a write-up of that evening: the sign-up screen, the settings, the security, and the wall I ran into when I actually tried to use it. Everything here is what happened to me, not what the marketing says should happen.

What Cursor actually is

If you've never touched it: Cursor is an AI code editor. Under the hood it's a fork of VS Code, so it looks and feels like the editor a lot of developers already live in. The difference is that the AI isn't a bolted-on plugin; it's wired into the core. You can ask it to write, explain, or refactor code, and it can read across your whole project instead of just the file you happen to have open.

It's built by a company called Anysphere, and it has grown quickly: over a million paying developers, with teams at places like Stripe, OpenAI, and Figma using it day to day. So this isn't a curiosity. It's one of the tools setting the pace in this space right now.

The sign-up, and a criticism that fell apart

The first screen offered the usual buttons (Google, GitHub, Apple) plus plain email. I went with email on purpose. Social logins are convenient, but I wanted to see how Cursor handled the boring, important parts of account security on its own.

After entering my address, it gave me two ways in: a one-time code sent to my inbox, or setting a password. Here's where I'll admit my bias. I chose the password route half-expecting to catch the app letting me straight in with nothing but a password, the kind of shortcut I could point to and use to knock the company for cutting a corner.

That's not what happened. Even after I set a password, it still asked for a one-time code from my email before it let me through. So the criticism I'd already lined up in my head quietly went away. Email verification is enforced either way, which is a small thing that a surprising number of products get lazy about. Credit where it's due.

Setting up the account

Once I was in, Cursor walked me through a short setup. It's quick, but a couple of these steps are worth slowing down on rather than clicking past.

Just for me, or for my team

First it asks how you'll use it: personal usage and individual settings, or collaboration features and shared context for a team. I picked the individual path.

Which role describes you

Then a row of tags: Software Engineer, Designer, Product Manager, Researcher, Data, Educator, Student, Other. Standard “help us understand you” segmentation. Harmless enough.

Start your Pro plan to continue

Next comes a full-screen nudge toward the paid plan: a bold Continue button, and a much quieter Skip for now underneath. You can absolutely skip it. It's just deliberately not the loud option.

Data sharing

The last step is the one I'd actually stop and read, so it gets its own section below.

The toggle that was already switched on

Cursor shows a “Share Data” screen with a single switch and a friendly line: let Cursor learn from how you code, chat, and work to make the AI better for everyone. The switch was already flipped on when I got there.

Then I hit the wall

With setup done, I landed on the dashboard. It cheerfully told me I'd used 0% of my usage, which, as I was about to learn, would stay 0% no matter what I tried.

I clicked into the agent to type my first prompt. The input box wouldn't take anything. A banner told me Cloud Agents requires a Pro account. Fair enough, I thought. Maybe cloud agents are the premium bit. So I went hunting for a free model I could actually run. I opened the model picker (Sonnet 4.6 was sitting right there) hoping for something lightweight that wouldn't need a subscription. Same wall.

Then I tried Bugbot, its code-review feature. Upgrade to Pro. Automations? Upgrade to Pro. Every door I pushed on in the web app led to the same place: a payment screen. For a first-time user trying to kick the tires, it's a frustrating loop.

So the wall is real, but it's specific. If you sign up expecting to try the product in your browser, you'll feel handed a locked showroom. If you download the desktop app, there's a genuine (if limited) free experience waiting. The onboarding just never makes that clear, and that gap is where a lot of the early frustration comes from. It's a signposting problem more than a bait-and-switch.

What it costs, and the catch

Cursor keeps the plans themselves simple: four tiers, from a free Hobby plan up to Enterprise, with a Monthly or Yearly toggle where yearly takes roughly 20% off. The part worth reading twice is the billing underneath the paid tiers.

Since a change in mid-2025, paid plans run on a credit system. Your monthly fee comes with a matching pool of usage credits. An “Auto” mode that picks a model for you is unlimited and doesn't touch that pool. But the moment you hand-pick a top-tier model, you start drawing it down, and heavy users can spend past their plan and pay overage on top. That 2025 change landed badly at the time. The company apologized and issued refunds, and the community is still a little wary of it.

Prices shown are the base tiers (Pro for Individual, Standard for Teams); yearly billing runs about 20% under monthly. Pro+ and Ultra are higher Individual tiers for heavier usage, and Bugbot runs on usage-based billing. Figures are from Cursor's pricing page in 2026 and can change, so check before you buy.

What it gets right, and where it grates

What other users say

My evening is one data point. Here's how the broader developer crowd tends to talk about it, boiled down.

How it broke down for me

Scored on what I could actually test from a fresh free account on the web, not on features locked behind the paywall.

My honest take after an evening with it

I like Cursor more than I expected to, and I trust its onboarding a little less than I'd like. Both of those things are true at once, and I don't think that's a contradiction.

The product itself is the easy part to recommend. If you write code for a living and you're already comfortable in VS Code, Cursor feels less like learning a new tool and more like your editor growing a very capable extra pair of hands. The codebase-wide understanding is the thing that actually earns the attention, and the autocomplete quietly does a lot of work in the background.

What I'd go in with eyes open about is everything around the editor, not the editor itself.

The free tier, at least on the web, is closer to a locked showroom than a test drive. You'll want the desktop app to feel the real thing. The billing rewards you for staying in Auto mode and can surprise you if you don't watch it. And a data-sharing toggle that's on by default is the kind of default I'd rather see flipped the other way.

So here's where I land. If you're ready to pay $20 a month (or $16 billed yearly) and you'll use it daily, it's a strong buy and probably worth the premium over cheaper options, as long as you keep half an eye on your credits. If you just want to try it for free, download the desktop app rather than judging it from the web dashboard, and don't expect the browser version to let you do much without a card.

I came in ready to be unimpressed. I left thinking it's a good tool wrapped in an onboarding that undersells its free side and quietly opts you into more than it needs to. Neither of those is a dealbreaker. Just know which one you're signing up for.