Cursor, reviewed
The editor that helped make AI part of everyday coding now belongs to SpaceX. What it does well, how its agents work, its pricing turns and its rivals.
Cursor did as much as any tool to make AI part of everyday programming. Anysphere, the company behind it, was founded in 2022 and built an editor with AI in every corner: completions that predict the next change, a chat that knows the whole codebase, and agents that work across many files at once. Its annualised revenue passed 2 billion dollars this spring, and on 14 August, after a deal valued at 60 billion dollars in stock, it became part of SpaceX.
It is also a tool whose ground has shifted twice in little more than a year, once on price and once on ownership. Here is what it does well, what changed, and how it compares.
What it does well
Cursor's strength has always been the feel of the editor. In our use, its completions guess not only the next line but the next edit, the chat understands the project it sits in, and switching between models from different labs takes a click. For experienced developers, that combination made AI part of the craft rather than a separate window.
Cursor describes its own path in one line: from “completing the next few lines of code” to building AI teammates that take on real work. In 2026 the second half of that sentence became the product.
From completion to agents
In January Cursor described running hundreds of agents on one project. In close to a week they wrote a web browser from scratch, more than a million lines across about 1'000 files. Planner agents broke the work into tasks, worker agents finished them, and a judge decided whether to go on. Cursor was frank about the limits: a large migration of its own code “still needs careful review”, and agents “occasionally run for far too long”.
In September that research reached customers. Projects, in beta for all users, hands a larger piece of work to a coordinator agent that writes no code itself. It gives tasks to agents in the cloud and on your machine, keeps context for months and keeps working when the laptop is closed. Two new bots follow a change from pull request to production. One reviews every change for security issues and proposes a fix. The other compares live signals after a release with the baseline before it and, depending on configuration, prepares a revert that waits for a person's approval.
Pricing, twice
In June 2025 Cursor moved its 20 dollar Pro plan from 500 requests a month to 20 dollars of usage at the model makers' API prices, with unlimited use only through Auto, which picks the model for you. Many users met charges they had not expected. On 4 July its cofounder and chief executive, Michael Truell, apologised, offered refunds for unexpected usage since 16 June and wrote: “We were not clear that ‘unlimited usage’ was only for Auto and not all other models.”
The second turn came with the new owner. Cursor's pricing page now sells subscriptions through X Subscription Corp. The 8 dollar Plus plan covers Grok models only, in a pool shared with Grok's own chat, image and voice products. Premium at 30 dollars, Super at 100 and Ultra at 200 include generous Grok usage but no allowance for third party models, which remain available at API list price. Plans have changed often, so check the page before you buy, and set a spending cap.
The competition
Cursor no longer has the field to itself. Claude Code, Anthropic's coding agent, passed 2.5 billion dollars in run rate revenue by March, Anthropic said. Cognition, the maker of Devin, raised 2 billion dollars in September at a valuation of 48 billion, with run rate revenue of 900 million. OpenAI has Codex, and GitHub's Copilot sits inside the platform where many companies already keep their code.
According to TechCrunch, investors said Cursor sold largely because it was severely short of computing power. The deal answers that: Cursor writes that it now has access to “the largest fleet of GPUs in the world”. It also ties the editor to one model maker, in a market where many developers pick a different model for each task.
Our verdict
Cursor is still one of the best places to write code with AI, and its agents are serious work, built with unusual openness about their limits. For a developer or a small team that wants the most polished editor, it is an easy recommendation.
For a company, the ground under the editor matters as much as the editor. Its owner builds Grok, its plans now lead with Grok, and, according to one report, a major supplier of competing models is about to leave. Keep your rules, prompts and tests in the repository rather than in the tool, cap the spending, and make sure your team could move to Claude Code, Codex or Copilot within a week.
We work the same way. In Werk, AI writes most of the code on a framework we built and trust, and people decide what to build, design how it feels and review every release. The tools change every quarter. The review does not.
