Cursor AI - a code editor with artificial intelligence - 2026

Published: 2026-07-29

Cursor (formerly known as Cursor AI) is an integrated development environment (IDE) with built-in artificial intelligence, developed by Anysphere (San Francisco, USA). The product is a fork of VS Code (Code — OSS) with deep integration of AI agents capable of autonomously editing code, analyzing projects, executing terminal commands, and interacting with external services through MCP (Model Context Protocol). As of mid-2026, Cursor holds approximately 31% of the AI developer tools market, making it the leader in usage share among professional developers.

History

Anysphere, Inc. was founded in 2022 by MIT students Michael Truell, Sualeh Asif, Arvid Lunnemark, and Aman Sanger. In October 2023, the startup announced an $8 million seed round led by the OpenAI Startup Fund, with participation from former GitHub CEO Nat Friedman and Dropbox co-founder Arash Ferdowsi.

In June 2025, Anysphere closed a $900 million Series C round at a $9.9 billion valuation led by Thrive Capital. By May 2026, the company's annual recurring revenue (ARR) exceeded $3 billion, and its valuation reached $29.3 billion.

On June 16, 2026, SpaceX announced the acquisition of Anysphere for $60 billion in stock. The deal, finalized days after SpaceX's historic IPO, places Cursor under the SpaceXAI subsidiary. The acquisition is expected to close in the third quarter of 2026.

Architecture

Core Engine

Cursor is built on Microsoft's open-source VS Code codebase (Code — OSS) but contains substantial modifications at the editor kernel level to support AI-first workflows. Unlike VS Code extensions, Cursor has native access to the AST tree, codebase indexing, and internal editor APIs.

Codebase Indexing

To power semantic search, Cursor builds a searchable index of the codebase when a project is opened. The process uses a Merkle tree to detect changes without reprocessing. A cryptographic hash is computed for each file; directory hashes are derived from child hashes. During synchronization, client and server hashes are compared — changed files are updated, unchanged ones are skipped.

When a file changes, Cursor splits it into syntactic chunks, which are converted into embeddings for semantic search. Embeddings are cached by chunk content — unchanged chunks use the cache, speeding up agent responses. According to internal Anysphere benchmarks, semantic search improves agent response accuracy by an average of 12.5%.

Privacy and Data Security

Cursor supports Privacy Mode, in which user code is never used for training by Cursor or any third-party model provider. Privacy Mode is enabled by default for Enterprise customers. Cursor maintains ZDR (Zero Data Retention) agreements with most model providers, guaranteeing that input and output data are not stored.

Cloud Agents are the only feature requiring code storage on Cursor's servers, as cloud agents need persistent repository access. This feature is optional and can be disabled.

Key Features

Feature	Description
AI Agent (Agent mode)	An autonomous AI agent that plans and executes multi-file changes: writes code, finds bugs, runs tests, and executes terminal commands. Available in both IDE and CLI.
Composer	Multi-file editing interface where Agent mode operates. Allows changes across dozens of files from a single command.
AI Chat	Built-in chat with AI that understands the full project context. Supports @-mentions of files, functions, and rules.
Tab Autocomplete	Predictive code completion widely regarded as the best in the industry. Supports multi-line completions with context from open files. Approximately 40% of Cursor users' code comes from accepting Tab suggestions. Unlimited on all paid plans.
Cmd+K (Inline Edit)	Inline prompt over the current selection. Allows rewriting, editing, or explaining selected code fragments.
Terminal Integration	The AI agent can execute terminal commands, analyze their output, and suggest fixes.
Codebase Understanding	Understanding of the architecture, file structure, dependencies, and logic of the entire project, not just the open file.
Refactoring	Rewriting large code sections while preserving logic, including automated refactoring that respects project style.
Debugging	The AI agent helps find and fix errors by analyzing stack traces, logs, and code.

Rules

Cursor supports four types of rules that govern AI agent behavior:

Always Apply — applied to every chat session;
Apply Intelligently — the agent decides when the rule is relevant;
Apply to Specific Files — triggered by file path patterns;
Apply Manually — invoked via @-mention (e.g., @my-rule).

Project rules are stored as .mdc files in the .cursor/rules directory and are version-controlled. Global rules are stored in Cursor settings. Enterprise customers have access to team-wide rules managed from the admin dashboard.

Skills

Skills are extensions for the AI agent that add specialized capabilities. For example, code review can be triggered via Bugbot directly from chat using the /review-bugbot or /review commands.

MCP (Model Context Protocol)

Cursor supports the Model Context Protocol (MCP) — an open protocol allowing AI agents to connect to external tools and services: databases, browsers, internal APIs, and file systems. MCP servers extend agent capabilities beyond the codebase.

Hooks

Hooks are scripts that execute before or after agent actions. They enable automated linter checks, formatting, test execution, and other tasks within CI/CD pipelines.

Bugbot (AI Code Review)

Bugbot is an automated pull request review tool. It analyzes the PR diff, leaves comments with problem descriptions and fix suggestions. It runs automatically on each PR update or manually via the cursor review (or bugbot run) command. The Enterprise version includes an API for triggering reviews and retrieving analytics.

Cloud Agents

Cloud Agents allow running AI agents on Cursor's cloud infrastructure. Each agent gets its own environment (virtual machine), can work with the repository, execute commands, and make changes. The key advantage is the ability to run agents in the background, even when the developer's computer is off. Multiple parallel agents can be launched, each with its own task.

CLI (Command Line Interface)

Cursor CLI is a command-line interface for working with AI agents outside the IDE. It installs with a single command:

curl https://cursor.com/install -fsS | bash

The CLI supports the same models and features as the desktop version, including Agent mode, file operations, and terminal commands. It can be used in GitHub Actions, CI/CD pipelines, and automation scripts.

Automations

Automations allow configuring AI agents to perform tasks automatically based on triggers — for example, on push to a repository or on a schedule.

Mobile

Cursor Mobile is a version of the AI agent for mobile devices, available via cursor.com/agents.

Worktree (/worktree)

The /worktree command creates a git worktree scoped to a specific task, allowing work on multiple branches simultaneously without context switching via stash.

Marketplace

Cursor Marketplace is a platform for publishing and sharing rules, skills, and plugins among users. Enterprise customers have access to an internal Marketplace with team-approved rules.

AI Models

Cursor supports a wide range of AI models from various providers, switchable directly within the editor:

Composer (Cursor's proprietary model)

Anysphere developed its own Composer model series based on open-source checkpoints. Key versions:

Composer 1.5 — February 2026;
Composer 2 — March 2026, built on Moonshot Kimi K2.5;
Composer 2.5 — May 2026. Substantial improvement in intelligence and behavior: better at sustained long-running tasks, follows complex instructions more reliably, improved communication style and effort calibration. Available in standard and fast variants. Pricing: $0.50/M input and $2.50/M output tokens (standard); $3.00/M and $15.00/M (Fast).

Third-party Models

Claude (Anthropic) — Claude Sonnet, Claude Opus 4.7, Claude Fable 5;
GPT (OpenAI) — GPT-4.1, GPT-5.6;
Gemini (Google DeepMind);
Grok (xAI) — integrated as Grok 4.5.

Model Switching

Users can select a model for each task. Different models are optimized for different scenarios: Composer 2.5 for everyday tasks and credit economy; Claude Opus for complex refactoring; GPT-5.6 for generation and analysis.

Pricing

As of July 2026, Cursor offers the following pricing plans:

Plan	Price	Key Features
Hobby (Free)	Free	Limited Agent requests, access to Composer, no credit card required
Pro	$20/mo	Extended Agent limits, generous Grok and Composer limits, frontier models, MCPs, skills, hooks, Cloud Agents, Bugbot (usage-based)
Pro+	—	Recommended for daily Agent users
Ultra	—	For power Agent users, maximum limits
Teams	$40/user/mo	Centralized team billing, team Marketplace, Bugbot code review, shared team context for Cloud Agents, usage analytics, team-wide Privacy Mode, SAML/OIDC SSO
Enterprise	Custom	Pooled usage, invoice/PO billing, SCIM, repository/model/MCP access controls, Audit Logs, AI code tracking API, priority support

Tab Autocomplete is unlimited on all paid plans. After the included request volume is consumed, usage-based billing is available.

Market Position

Market Share (February 2026)

Tool	Usage Share	Change from March 2025
Cursor	31%	−3% (was 34%)
GitHub Copilot	24%	−4% (was 28%)
Claude Code	22%	+10% (was 12%)
Windsurf (Codeium)	8%	+5% (was ~3%)
JetBrains AI	4%	New entrant
Replit AI	3%	−2%
Other / None	8%	−6%

Cursor remains the market leader, though its share has decreased slightly due to the rise of Claude Code. 53% of developers use two or more AI tools; the most popular combination is Cursor + Claude Code (14% of respondents). Average daily Cursor usage is 4.2 hours.

Comparison with Alternatives

Dimension	Cursor	GitHub Copilot	Claude Code
Interface	AI-native IDE (VS Code fork)	Extension for VS Code, JetBrains, Neovim	Terminal / CLI, no IDE required
Agent Autonomy	Medium-high (plans + applies with diffs)	Medium (proposes changes, requires manual application)	High (autonomous edits + shell commands)
Model Flexibility	Multiple providers, in-editor switching	Multi-model selector (widest provider range)	Claude models only
Starting Price	Free; Pro $20/mo	Free; Pro $10/mo	API (usage-based) or Max subscription
Codebase Understanding	Full (Merkle tree indexing)	Limited (1-2 file context)	Full (project analysis via filesystem)
Cloud Agents	Yes (cloud VMs for background work)	No	No

Enterprise Features

Cursor Enterprise includes:

Pooled usage — shared request pool across the team;
SCIM — user management via System for Cross-domain Identity Management;
SAML/OIDC SSO — single sign-on;
Access controls for repositories, models, and MCP servers;
Audit Logs;
Service accounts for automation;
AI code tracking API — tracking the proportion of AI-generated code;
Auto-run, browser, and network controls;
Team Marketplace — internal rules and skills publishing;
Priority support and dedicated account management.

Community and Ecosystem

Cursor has an active developer community, a forum at forum.cursor.com, regular workshops, and a beta testing program. The company publishes open benchmarks (Evals at cursor.com/evals) and technical blogs about model development and architecture. The Cursor Rules concept was adapted by the community into the open AGENTS.md standard, now supported by many AI tools.

SpaceX Acquisition

On June 16, 2026, days after its record-breaking IPO, SpaceX announced the acquisition of Anysphere (Cursor's parent company) for $60 billion in stock. The deal includes termination fees: $10 billion under certain conditions, $4 billion if blocked by antitrust regulators. Cursor will operate under the SpaceXAI subsidiary. Expected closing date is the third quarter of 2026. According to Reuters, SpaceX had been considering the acquisition since April 2026, alternatively offering a $10 billion partnership.

Criticism and Limitations

Dependency on cloud AI models: Agent mode and semantic search require an internet connection.
Cost: heavy Agent usage on large projects can lead to significant expenses on usage-based plans.
Ecosystem lock-in: while Cursor is based on VS Code, not all VS Code extensions are fully compatible.
Competition: rapid growth of Claude Code (from 12% to 22% in one year) puts pressure on Cursor's market share.

Resume

Cursor AI is an AI-native integrated development environment, leading the AI developer tools market with approximately 31% share. The product combines a powerful AI agent capable of autonomously performing tasks from code writing to refactoring and debugging, with a familiar VS Code-based interface. Key innovations include the proprietary Composer 2.5 model, Cloud Agents, automated code review (Bugbot), a system of rules and skills, and MCP support. The $60 billion acquisition by SpaceX in 2026 confirms the strategic importance of AI development tools. Cursor remains the primary choice for developers requiring deep AI integration with the ability to work on entire projects rather than individual files.

Links

×

← В главную