Cursor Review

Founded 2023, and SpaceX completed its acquisition of Cursor in August 2026. The changelog is active, the pricing page is clean, and they're not playing games with seat traps. VS Code foundation means switching costs are near zero in both directions, which is exactly what you want in a category still sorting itself out.

Two things concern me. One: Teams is pitched as self-serve, with priority support held back for Enterprise, which isn't confidence-inspiring for a $40/seat Teams tier. Two: the free plan caps you at 'limited Agent requests' — 'limited' is doing real work in that sentence. Engineers will hit the ceiling and expense Pro before you've standardized anything.

The codebase indexing feature is the real differentiator. GitHub Copilot now ships agent mode on every plan, so agents alone don't set Cursor apart; an agent that works from your full indexed project graph does. That's a different category of useful for teams with complex dependency trees.

The tradeoff nobody discusses: you're routing proprietary code through AI models. The .cursorignore file helps, but it requires discipline your least careful engineer won't exercise. Worth a privacy audit before Teams rollout.

Founded 2023, shipping a changelog, $20/month Pro tier with frontier models from OpenAI and Anthropic — that's a credible early-stage infrastructure story, not a demo. Building on VS Code's proven architecture was the right call. It means the extension ecosystem comes for free, keybinding muscle memory transfers, and enterprise security teams aren't staring at an unfamiliar attack surface. The .cursorignore file support for privacy-scoped analysis tells me someone thought about enterprise procurement before launch, not after.

The codebase indexing feature is where Cursor separates itself from GitHub Copilot, which now has agent mode on every plan but lists organization-wide codebase indexing as a Copilot Enterprise feature. Full-project semantic awareness changes what you can ask the AI to do — cross-file refactoring, dependency-aware generation, architectural queries. That's a meaningfully higher craft ceiling than autocomplete-with-benefits.

The strategic risk lives in the model layer. At $20/month, Pro splits its included usage into two pools: Cursor's own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5) and an Other Models pool billed at OpenAI, Anthropic and Google API rates. If those providers change pricing or API terms, that half of Cursor's unit economics moves without warning. If we're a 50-engineer org standardizing on Teams at $40/seat per month, that dependency is a procurement conversation we need to have annually, not once.

Real-time collaborative editing isn't something I could confirm — teams share AI-generated code through Git, which is fine but puts Cursor behind where Replit and some emerging multiplayer IDEs are heading. That's not a blocker today. In year two of a growing team, it might be.

Pricing page shows every self-serve tier without a sales call: Hobby $0, Pro $20, Pro+ $60, Ultra $200, Teams $40 or $120/seat/month. That's procurement-friendly. No hidden SSO line item, which GitHub Copilot Business at $19/seat also avoids, and Copilot Business offers Anthropic and OpenAI models too. So Cursor's $40 Teams seat has to earn its premium on agent depth, not on model access.

50-seat team on Pro: $20 × 50 × 12 = $12K/year. Add standard 30% seat creep by year 3, you're at $15.6K annually, roughly $41K over 3 years. Upgrade half those seats to Teams for admin controls and analytics: the blended rate pushes the 3-year total closer to $62K. Not alarming. Predictable, which matters more than cheap.

The real exposure is on-demand usage. Pro includes a fixed monthly usage allowance per seat; past it, requests bill at published per-model API rates, and Teams adds $0.25 per million tokens on third-party models. That's the number to model before signing — not the sticker. The free tier's limit is just 'limited Agent requests', with no hard number. Vagueness on a public pricing page is a yellow flag worth clarifying before committing.

Founded 2023, built on VS Code's foundation. Migration story is real — extensions, keybindings, settings all port over. That's not marketing copy, that's an actual engineering decision that matters. You don't lose your muscle memory on day one, which is usually where AI tool adoption dies. The .cursorignore file for excluding sensitive paths is the kind of thing a developer shipped, not a PM.

The Hobby tier is where teams get burned. 'Limited Agent requests' sounds workable until you're pairing AI on a refactor-heavy sprint. The pricing page never says how limited — no request count, no completion count, just the word. That vagueness in public documentation is a flag. Stripe doesn't have that problem. Neither does GitHub Copilot's pricing page.

Codebase indexing — where the AI understands your project's dependency graph, not just the open file — is the feature that separates Cursor from a glorified autocomplete. GitHub Copilot now has agent mode on every plan, but it lists organization-wide codebase indexing as a Copilot Enterprise feature. Codebase-aware suggestions in long-lived repos with complex import chains is where this earns its $20/month.

Priority support is Enterprise-only. The Cloud Agents API is still in public beta. That's a real constraint for teams wanting to pipe Cursor data into internal tooling or incident workflows. Teams at $40/seat per month adds centralized billing and usage analytics, which is fine, but on-demand charges once each seat's included usage runs dry will still sting on active teams.

Founded in 2023, Cursor made a smart bet: don't build a new editor, just take VS Code and make it think. That decision pays off immediately. Your extensions migrate over, your keybindings are there, you're not relearning muscle memory. The first ten minutes feel like welcome, not homework. That's rarer than it should be.

The codebase indexing is the feature that separates this from GitHub Copilot slapped into an editor. It's not just autocomplete — the AI chat actually knows your project structure, your dependencies, your patterns. Ask it why something is wired the way it is, and it can actually tell you. That's genuinely useful at 3pm when you've forgotten why you wrote something three weeks ago.

Here's the honest part though: the Hobby plan gives you 'limited Agent requests' and nothing more specific. Limited. You'll feel that ceiling faster than you expect, and the step to $20/month Pro is a real commitment. No gradual middle ground. Also, the main editor is a desktop app — Mac, Windows, Linux. The iOS app lets you start agents and review and merge PRs from your iPhone, but there's no editor or terminal in it, and Android has no native app yet.

Three months in, the question becomes whether the AI suggestions stay sharp or start feeling repetitive. The changelog shows active development, which matters. Products like this either compound in value or plateau hard.

'We are under no obligation to provide support for the Service.' Section 17.4 of Cursor's terms. Teams seats run $40 and $120 per user a month, and priority support still sits on Enterprise. Category norm, maybe. Worth reading before the first invoice anyway.

Then the models page. SpaceX closed its acquisition on August 14, 2026. Pro, Pro+ and Ultra split usage into two pools: Grok 4.7, 4.6, 4.5 and Composer 2.5 get significantly more included usage, while OpenAI, Anthropic and Google models bill at API price. Teams adds a $0.25 per million-token Cursor Token Rate on third-party models. Grok and Composer are exempt. Nothing hidden. But the pricing leans one way.

Fair's fair. The changelog shipped Rollouts and Security Review on September 23, and leaving still means going back to VS Code. However, the Cloud Agents API v1 is public beta — 'APIs may change before general availability.' GitHub Copilot has agent mode on every plan, Free included, so agents alone aren't a moat. Build nothing on that API you can't rewrite.