Cursor’s bet, paid off, was that an AI-first editor would draw in developers willing to leave VS Code if the AI experience was meaningfully better than VS Code plus an extension. The fork is binary-compatible with most VS Code extensions, which keeps the switching cost low for teams that aren’t deeply invested in unusual tooling. Once the editor is open, the Agent panel — and increasingly the cloud-agent surface in Pro+ and Ultra — is where most users spend their time.
The reason Cursor’s pricing ladder has four individual tiers is that “AI usage” varies wildly across developers in a way that earlier coding-tool pricing models didn’t have to handle. A backend engineer running a single agent task per day will exhaust Pro’s allowance comfortably; a frontend engineer iterating on a Next.js app with the cloud agent open all day will hit Pro+‘s 3× envelope and benefit from Ultra’s 20×. The published rule is “extra usage at API pricing” once an allowance is depleted, which means the question for a buyer is “what’s the monthly API cost equivalent of our actual usage, and at which tier does the subscription pay for itself?” That calculation favours Pro+ for most heavy individual users and the $40/seat Teams plan for lean groups that want pooled controls without Enterprise pricing.
What the editor lock-in actually costs
Cursor inherits VS Code’s extension marketplace, so most of the surface a developer relies on — language servers, theme, debugger, format-on-save — works without modification. The lock-in lives in two places. First, Cursor’s release cadence diverges from upstream VS Code, so teams that pin specific VS Code versions for compliance reasons need to decide which release line they trust. Second, organisations standardised on JetBrains, Visual Studio, or Vim cannot meaningfully use Cursor; “let one team use Cursor, the rest stay on Copilot” is workable for a 20-person company and untenable for a 200-person one.
The other adoption friction is data flow. Cursor’s broad model menu — OpenAI, Anthropic, Google, plus the editor’s own indexing services — means that a regulated organisation has multiple vendor approvals to obtain, not one. The Privacy Mode setting reduces but does not eliminate this fan-out. For most lean teams this is a non-issue; for healthcare, legal, and finance buyers, it’s the conversation that decides the procurement.
Where it earns the seat price
The clearest “yes” signal we see in real teams using Cursor is when a developer reports that they’re shipping work they would not have attempted before — a refactor that would have taken a week as an afternoon’s task, an experimental feature that would have been put off entirely. At that point the question of “does $20/seat pay for itself” stops being a calculation and becomes obvious. The follow-on question — “Pro, Pro+, or Ultra” — is the one most teams need actual usage data to answer; the published 1×/3×/20× multipliers map almost directly to “occasional”, “primary”, and “always-on” agent usage profiles.