Cursor 1.0 turns its AI into an agent: PR review, background coding, one-click MCP

By 2025 most AI coding tools still autocompleted or chatted; the human did the editing, reviewing and context-gathering. Cursor 1.0, the first major release since the 2023 launch, bet that the assistant should act, not just suggest — and that the editor itself should be built around that.

The editor's AI stops autocompleting and starts doing: it reviews pull requests, installs its own tools, and codes in the background while the developer keeps writing.

Cursor 1.0 opened Background Agent to every user (remote, cloud-based coding launched from a cloud icon or Cmd/Ctrl+E), added Bugbot to automatically review GitHub pull requests and leave 'Fix in Cursor' comments, one-click MCP server installs with OAuth, Memories that persist facts across conversations, and editing inside Jupyter notebooks.

The release moved agent features out of early access into the default product, and the agent — not the autocomplete — became the center of gravity for the AI-editor category. At the same time Anysphere reported $500M in annualized revenue, and by November 2025 it had passed $1B ARR with a $29.3B valuation.

Cursor began in 2022 as the bet that the code editor itself should be rebuilt around AI, not just get an AI sidebar. CEO Michael Truell describes the product as 'the best way to code with AI' — a souped-up word processor where the tool predicts the next twenty minutes of your work and, increasingly, takes delegated tasks the way a pair programmer would.

Cursor 1.0, released June 4, 2025, was the first major version since launch and turned that philosophy into concrete product: Bugbot automatically reviews pull requests on GitHub and leaves comments you can fix with one click; Background Agent, previously early access, opened to all users so coding work happens remotely while the developer keeps editing; one-click MCP installs and OAuth let the editor wire itself to external tools; Memories persist facts between conversations.

The release marked the moment the AI agent became the product rather than a feature. Competitors from GitHub Copilot to Windsurf moved the same direction — agents, PR review and background work — within months. The Verge, calling Cursor 'one of the most popular and fastest-growing AI products in the world,' interviewed Truell in August 2025 as the category raced to adopt agentic coding.

Business results followed the product shift: Anysphere reported $500M in annualized revenue when the round around the 1.0 launch valued it at $9.9B, and by November 2025 its Series D valued the company at $29.3B with annualized revenue past $1B.

Autocomplete leaves the human with all the context-gathering and reviewing; an agent takes over whole tasks end to end.

PR review inside the editor closes the loop: the AI that wrote code also checks it, and the fix is one click away.

Background agents let work happen outside the human's attention, turning the editor into a manager of parallel helpers.

One-click MCP turned tool integration from a setup chore into a product feature, lowering the cost of trying the agent.

When a tool's AI stops suggesting and starts acting, the product question shifts from model quality to what the tool can do on its own — and what the human still needs to see.

Cursor 1.0 shipped on June 4, 2025. Six months later Anysphere closed its third round in under a year: a $900M Series C at a $9.9B valuation with $500M annualized revenue (Thrive Capital, a16z, Accel, DST), then a $2.3B Series D led by Accel and Coatue — with Nvidia and Google as new investors — valuing the company at $29.3B as revenue passed $1B a year. Millions of developers use Cursor, the team grew past 300 people, and in late October 2025 the company shipped its first self-developed coding model, Composer.