Cursor

Cursor is an AI coding agent built by Anysphere, Inc. that combines a familiar code editor with autonomous AI agents capable of planning, writing, debugging, and reviewing software end to end. It runs across desktop, CLI, web, mobile (iOS), and integrations like Slack, GitHub, and JetBrains IDEs. The product is actively developed and as of version 3.9 (June 2026) includes a native iOS mobile app, cloud agents, automations, and a marketplace for community-built plugins.

What It Is

Cursor is a coding agent platform — not just an autocomplete tool — that lets developers delegate implementation tasks to AI while staying in control of higher-level decisions. It ships as a desktop editor (built on VS Code foundations), a CLI, a web/mobile agent dashboard, and integrations for existing tools. Agents can read files, run terminal commands, search the web, browse the UI, and submit pull requests autonomously. The core interaction model spans a spectrum from inline Tab completions and targeted Cmd+K edits to fully autonomous multi-agent runs that work in parallel across a codebase.

How the Autonomy Spectrum Works

Cursor's product page describes three primary interaction modes:

• Tab completion — context-aware inline suggestions as you type
• Inline edit (Cmd+K) — targeted rewrites of selected code
• Agent mode — full autonomous execution: the agent plans, asks clarifying questions, edits files, runs shell commands, takes screenshots, and iterates until the task is done

A Plan mode lets agents ask clarifying questions and build a structured task list before executing, reducing wasted work on complex features. A Debug mode instruments code with real execution data to pinpoint root causes. A Design mode lets agents visually edit UI by selecting elements directly in a browser preview.

Platform and Integration Breadth

Cursor runs everywhere developers work:

• Desktop — a full editor for manual-to-agentic coding
• CLI — cursor agent runs in any terminal, script, or CI pipeline
• Web & Mobile — cloud agents accessible from a browser or the iOS app (launched June 2026)
• Slack — agents respond to @cursor mentions and open PRs directly from channels
• GitHub / GitLab / Azure DevOps / Bitbucket — PR review via Bugbot, which automatically scans new PRs for bugs
• JetBrains IDEs — via ACP-compliant agent integration
• Linear, Notion, Jira, Microsoft Teams — listed as supported integrations in docs

MCP (Model Context Protocol) support lets teams connect external tools and data sources like Figma directly to the agent. A Skills system lets teams add domain-specific prompts and code that agents can discover and invoke.

Model Access and Codebase Intelligence

Cursor provides access to frontier models from Anthropic (Claude Opus 4.8, Claude Sonnet 5, Claude Fable 5), OpenAI (GPT-5.5, GPT-5.3 Codex), Google (Gemini 3.1 Pro, Gemini 3.5 Flash), xAI (Grok 4.3), and Cursor's own Composer 2.5 model. An "Auto" mode selects the best model per task. Subagents can run in parallel, each using a different model optimized for its subtask.

Codebase indexing uses a custom embedding model for semantic search across large repos. The enterprise page states Cursor is architected to handle codebases with millions of lines across hundreds of thousands of files, and a 2026 research post describes secure reuse of a teammate's existing index to cut time-to-first-query from hours to seconds on the largest repos.

Enterprise Adoption and Scale Signals

The enterprise page publishes several vendor-claimed figures: Cursor says 64% of Fortune 500 companies use the product, over 50,000 enterprises have chosen it, and more than 100 million lines of enterprise code are written per day with Cursor. Published customer quotes include Jensen Huang (NVIDIA) stating all ~40,000 NVIDIA engineers are assisted by AI via Cursor, Patrick Collison (Stripe) describing growth from hundreds to thousands of enthusiastic Stripe employees, Brian Armstrong (Coinbase) noting every Coinbase engineer had utilized Cursor by February 2025, and James Reggio (Brex CTO) reporting more than 70% of Brex engineers use Cursor. The enterprise page also claims that in head-to-head evaluations, 93% of engineers select Cursor as their preferred AI coding tool — a vendor-published figure.

Enterprise features include SAML/OIDC SSO, SCIM provisioning, pooled usage, audit logs, repository/model/MCP access controls, auto-run and network controls, SOC 2 Type II certification, AES-256 encryption at rest, and TLS 1.2+ in transit.

Update: Version 3.9 and Recent Releases

The changelog shows active, rapid release cadence in 2026:

• 3.9 (June 29, 2026) — Cursor Mobile App for iOS launched, enabling cloud agents from a phone
• 3.9 (June 22, 2026) — Customize Cursor settings
• 3.8 (June 18, 2026) — Improvements to Cursor Automations
• 3.7 (June 17, 2026) — Cloud Environment Setup and Cloud Subagents in Agents Window

Research posts from early 2026 cover secure codebase indexing, scaling long-running autonomous coding agents, and dynamic context discovery. A technical report on Composer 2 was published in March 2026. The product direction signal is clear: Cursor is moving toward always-on, multi-agent, cross-surface autonomous development with cloud infrastructure at the center.