Cursor is the AI-native integrated development environment that has, over the course of 2024 through 2026, become the most-adopted AI coding tool in the professional developer market. The product is technically a fork of Microsoft’s Visual Studio Code with substantially deeper AI integration layered into the IDE rather than added as a plugin. The company behind it, Anysphere, has gone from a relatively unknown startup to one of the most valuable AI companies in the world in roughly two years, with a customer base that includes a substantial fraction of professional software developers and a strategic position that lets them increasingly shape what AI-assisted development looks like in practice.
This piece is the foundational pillar on Cursor. We cover the company background, what Cursor actually is and how it differs from the alternatives, the core features that distinguish the product, the model strategy (which has shifted meaningfully through 2025 and 2026), the pricing structure, the customer profile, the competitive landscape, and the broader strategic story of why Cursor matters beyond just being a popular IDE. Subsequent posts will go deep on specific Cursor features, including the Composer model lineage and the Cursor 2.5 release that prompted this pillar.
The short version is that Cursor occupies a position no other AI coding tool has quite achieved: the familiar-IDE surface that adopters can use immediately without changing their workflow, combined with deeper AI integration than any IDE plugin can provide, combined with a model strategy that is becoming independent of any single frontier-model provider. The combination is the reason Cursor has taken share from GitHub Copilot, why it has not been displaced by Claude Code or Antigravity, and why Anysphere has been able to raise at valuations that would seem aggressive for a tool category that some observers initially dismissed as a wrapper around someone else’s models.
Anysphere and the founding story
Cursor is the product of Anysphere, a company founded in 2022 by four MIT graduates: Michael Truell, Sualeh Asif, Arvid Lunnemark, and Aman Sanger. The team had been thinking about how AI would change software development for several years before the GPT-3.5 and ChatGPT moment in late 2022 made the question urgent. The founding insight was that the IDE was the right surface for AI-assisted development, not a chatbot or a separate tool, because the IDE is where developers actually work and where the context for useful AI assistance already exists.
The early Cursor product launched in early 2023 as a Visual Studio Code fork with AI features built in. The choice to fork VS Code rather than build a new IDE from scratch was strategic. VS Code is the dominant developer IDE by a wide margin, with a deep extension ecosystem and an installed base measured in tens of millions of developers. Forking VS Code meant Cursor users got their familiar editor, their existing extensions, their existing keybindings, and the development experience they were already used to, plus the AI features. The friction of adopting Cursor was approximately zero for a VS Code user.
The product gained traction quickly through 2023 and 2024 as the AI features matured. By early 2025, Cursor had become the visible leader in the AI IDE category, with substantial adoption among professional engineers at startups and increasingly at larger organizations. The company’s funding trajectory tracked the adoption: from a seed round in 2022 through Series A in early 2024 at a roughly billion-dollar valuation, and through subsequent rounds at increasingly large valuations through 2025 and into 2026. The company has been one of the fastest-growing companies in the broader AI ecosystem by both revenue and valuation.
The team has remained small relative to the valuation. As of mid-2026, Anysphere’s headcount is in the low hundreds, which is small for a company of this scale and is consistent with the broader pattern of AI-native companies operating with leaner teams than the traditional software-company benchmarks suggest.
What Cursor actually is
Cursor is a desktop application that looks and feels like VS Code because it is a fork of VS Code with the AI features added at the IDE layer rather than as extensions. When you open Cursor, you see the familiar VS Code interface: the file explorer sidebar, the editor tabs, the integrated terminal, the bottom status bar, the extensions marketplace. Most VS Code extensions work in Cursor without modification, including the major language servers, formatters, debuggers, and theme packs.
What Cursor adds, on top of the VS Code base, is a set of AI features that have access to deeper context than a plugin could obtain. The core AI surfaces:
Tab autocomplete is the inline AI suggestion that appears as you type. As you write code, Cursor’s autocomplete model continuously predicts what you are likely to type next and shows the suggestion as ghost text that you can accept by pressing Tab. The autocomplete is not just next-token prediction; it can suggest multi-line edits, including edits that span the existing surrounding code (deleting parts, inserting parts, rearranging parts). The autocomplete model has been one of Cursor’s most-iterated surfaces and is one of the features most often credited as the reason developers stay on Cursor after first adopting it.
Cmd-K inline editing lets you select code, press Cmd-K (or Ctrl-K on non-Mac systems), describe the change you want in natural language, and have Cursor make the change directly in the editor. This is the surface for small, focused edits where you know what you want but would rather describe it than type it. The inline editor preserves the surrounding context, makes the change conservatively, and lets you accept or reject the result with a single keystroke.
Composer is the multi-file agent. Cmd-I (or Ctrl-I) opens the Composer surface, where you can describe a more substantial change that may span multiple files. Composer plans the change, makes edits across the affected files, runs commands in the terminal as needed, and reports back with what was changed. This is the surface that most directly competes with Claude Code, Antigravity, and the other agentic coding tools. The Composer feature has had its own model lineage (Composer 1 in October 2025, Composer 2 in March 2026, Composer 2.5 in May 2026) and is the surface where Cursor is increasingly using its own custom-trained models rather than only routing to frontier providers.
Chat is the conversational surface for asking questions about your codebase, getting explanations of code, and having longer-running discussions about implementation approaches. Chat has access to the full codebase as context (through Cursor’s indexing layer), to the open editor tabs, to the integrated terminal, and to whatever documentation or web sources you grant it access to.
These four AI surfaces (Tab, Cmd-K, Composer, Chat) cover the spectrum of AI-assisted development from "subtle inline help" to "full agentic multi-file work." Each surface is appropriate for a different scale of task. The integrated nature of having all four in a single IDE is the practical reason Cursor’s user experience is meaningfully better than the alternative of stitching together separate tools for each.
The model strategy
Cursor’s model strategy has shifted meaningfully through 2025 and 2026, and understanding the shift is important for understanding where Cursor is going.
In the early years, Cursor was primarily a routing layer in front of frontier models from Anthropic and OpenAI. The user selected which underlying model to use (Claude Sonnet, Claude Opus, GPT-4, GPT-5), and Cursor’s value-add was the IDE integration, the prompt engineering, and the agentic harness around the model. The model providers got the credit (and the token costs) for the actual capability; Cursor got the credit for the developer experience.
This worked but had two strategic limitations. The first was vendor dependency: Cursor’s product quality was tied to the underlying models, which Cursor did not control. The second was cost structure: a substantial fraction of Cursor’s revenue went directly to the model providers as token costs, which limited Cursor’s gross margins.
Beginning in 2025, Cursor started training its own models. The first major in-house model was Composer 1, released October 29, 2025. Composer 1 was followed by Composer 2 in March 2026 and Composer 2.5 in May 2026. The Composer model lineage is Cursor’s own coding-agent model, used inside the Composer feature for the agentic multi-file work.
The Composer models are not currently positioned as Cursor’s only model option. The IDE still lets users select frontier models from Anthropic, OpenAI, and Google for the various AI surfaces. The Composer model is the recommended default for the Composer agentic feature, particularly for cost-sensitive workflows where the cost-per-task economics favor the in-house model over the more expensive frontier alternatives.
The strategic direction the Composer model lineage represents is clearer in retrospect than it was at any single release moment. Cursor is moving toward foundation-model independence, with the in-house Composer models providing a fallback to ensure Cursor’s product continues to function (and to maintain margins) even as the frontier-model providers’ pricing, policies, or availability shift. This is the same strategic pattern that other AI-application companies have been pursuing, and it is the right pattern for any company whose business depends on AI capability that it does not control.
The most recent strategic move in this direction is Cursor’s announced partnership with SpaceXAI to train a from-scratch Cursor foundation model using approximately ten times more compute than Composer 2.5 used. The Colossus 2 cluster’s million-H100-equivalent capacity is the compute scale that this partnership unlocks. The from-scratch model would represent a meaningful step beyond the open-source-base fine-tuning approach that Composer 1 through 2.5 used.
Pricing
Cursor’s pricing structure is straightforward:
Free tier includes the IDE with limited AI usage (a capped number of AI requests per month). The Free tier is enough for individual developers to evaluate Cursor and use it for light AI-assisted work.
Pro tier at $20 per month per user includes unlimited Tab autocomplete, generous quotas on the more expensive AI surfaces (Cmd-K, Composer, Chat), and access to the full range of supported underlying models. Pro is the right tier for most individual developers who use Cursor regularly.
Business tier at $40 per month per user adds team management features (centralized billing, SSO, admin controls), higher quotas on the expensive AI surfaces, and the ability for organizations to manage their AI spend at the team level rather than the individual level.
Enterprise tier with custom pricing adds enterprise security features (audit logging, data-residency controls, custom retention), enterprise SSO, dedicated support, and the ability to negotiate underlying model pricing for high-volume customers. Enterprise is the tier for large organizations with formal AI procurement requirements.
The pricing has been stable since early 2024, though the value at each tier has increased substantially as Cursor’s AI features have improved. The Pro tier in particular has become substantially more capable over time without the price increasing.
A separate pricing dimension is the cost of the underlying frontier-model usage. When a Cursor user invokes a Claude or GPT model through the Cursor surface, the token usage is charged either to Cursor (in which case it counts against the user’s plan quotas) or to the user’s own model-provider account (if the user has configured Cursor to use their own API keys). The dual-billing model is meaningful for organizations that want to use their own negotiated model pricing rather than Cursor’s.
The customer profile
Cursor’s customer base in mid-2026 includes a substantial fraction of professional developers who do AI-assisted coding regularly. The pattern that has emerged:
Individual professional developers at startups and small companies typically adopt Cursor on the Pro tier through individual signups. The cost is low enough relative to a developer’s salary that the productivity gain justifies the personal expense even when the employer is not paying. This pattern accounts for the largest user-count share of Cursor’s base.
Engineering teams at small-to-mid startups typically adopt Cursor on the Pro or Business tier through a team-led rollout. The decision is often made by an engineering manager or technical lead who has personally adopted Cursor and wants their team to have the same productivity gain. The Business tier’s centralized billing and admin controls make the team rollout straightforward.
Engineering organizations at larger companies typically adopt Cursor on the Business or Enterprise tier through a more formal procurement process. The decision involves security review, legal review, and integration with the company’s existing identity and security infrastructure. The Enterprise tier handles the requirements of this segment.
Cursor’s largest enterprise customers include several major technology companies and an increasing number of Fortune 500 enterprises in non-technology industries. The specific customer logos are typically only disclosed in case studies, but the broad pattern is that Cursor has been moving up-market through 2025 and 2026 from its original startup-developer base into the enterprise segment.
The customer profile has been a meaningful concern for some observers because Cursor’s strongest competitive position is with the developer-led-adoption pattern rather than with formal enterprise procurement. The move up-market into enterprise requires the company to invest in capabilities (security, compliance, support) that the original startup-developer model did not need. Anysphere has been making those investments through 2025 and 2026.
The competitive landscape
The AI coding tool category in mid-2026 has several distinct competitors and several distinct competitive positions:
GitHub Copilot (Microsoft) remains the most-installed AI coding tool by raw user count because of the GitHub distribution channel and the integration with Visual Studio and VS Code. Copilot’s strongest position is the convenience of being already-installed for developers in the Microsoft ecosystem. Cursor’s competitive advantage versus Copilot is the deeper AI integration (Composer in particular) and the model flexibility. Many developers use both, with Copilot for the tasks that fit its surface and Cursor for the tasks that fit Cursor’s.
Claude Code (Anthropic) is the agentic-coding tool from Anthropic, distinct from Cursor in that it does not include an IDE but operates as a CLI tool that integrates with the developer’s existing editor. Claude Code’s strongest position is the deep Anthropic-model capability and the agentic harness designed specifically around Claude. Cursor’s competitive advantage versus Claude Code is the IDE integration; Claude Code’s competitive advantage is that you can use it inside any editor including Cursor itself.
Google Antigravity is the agentic IDE from Google, launched in March 2026. Antigravity competes most directly with Cursor on the IDE-plus-agent combination. Antigravity’s strongest position is the integration with Gemini models and with Google Cloud deployment paths. Cursor’s competitive advantage is the broader model support (including Claude and GPT models, not just Gemini) and the VS Code compatibility.
OpenAI Codex is OpenAI’s agentic coding tool, accessible through the OpenAI API and through Codex CLI. Codex competes for the same agentic-coding workloads but does not include an IDE surface. Cursor’s competitive advantage versus Codex is the integrated IDE; Codex’s competitive advantage is access to the latest OpenAI models.
Windsurf (formerly Codeium) is the second AI-native IDE competitor, also a VS Code fork with deeper AI integration. Windsurf was acquired by Cognition in mid-2025 and has been undergoing strategic shifts since. The market position has been less clear through 2026 than in earlier years.
Replit Agent is Replit’s hosted agentic coding surface, distinct from the IDE-on-desktop pattern because Replit hosts the entire development environment. Replit Agent’s strongest position is the fully-hosted, no-local-setup story for users who want everything in the browser.
Augment Code is the agentic coding tool from Augment, focused specifically on enterprise codebases and the patterns of working in very large existing codebases (millions of lines of code, many years of history). Augment’s strongest position is the codebase-scale story.
JetBrains AI Assistant is the AI integration for the JetBrains IDE family (IntelliJ, PyCharm, GoLand, and similar). JetBrains’ strongest position is the IntelliJ-family installed base, which is the dominant IDE for Java, Kotlin, and several other language ecosystems. Cursor competes less directly with JetBrains because the user bases overlap less.
The market is genuinely competitive and the right tool for a given developer depends on the specific workflow, the language ecosystem, the model preferences, and the organizational constraints. Cursor’s strongest position is the combination of familiar IDE plus deep AI integration plus model flexibility, which fits the largest single segment of the market (professional developers in VS-Code-compatible language ecosystems) better than any single competitor does.
The strategic story
The broader strategic significance of Cursor goes beyond being a popular IDE. Three threads worth understanding:
The IDE as the right AI surface. Cursor’s success has settled an architectural question that was open in 2022 and 2023: where should AI-assisted development live? Various proposals existed (chatbot, separate web tool, code review surface, terminal CLI). The IDE has won, both because that is where developers spend their time and because that is where the context for useful AI assistance is most readily available. Most subsequent AI coding tools have converged on either the IDE pattern (Cursor, Antigravity, Windsurf) or the IDE-integrated pattern (Claude Code, Codex, Augment, JetBrains AI). The IDE-or-IDE-integrated pattern is now the dominant architecture for AI coding tools.
The verticalization of AI applications. Cursor’s move from "routing layer for frontier models" to "custom models for specific use cases" is one example of a broader pattern. AI-application companies are increasingly investing in their own model capability rather than depending entirely on frontier providers. The economics of this move depend on the specific application’s volume and the cost gap between routing and self-training, but the strategic logic is consistent across the category. Cursor’s Composer model lineage is the most-discussed example because Cursor has been transparent about it; many other companies are pursuing similar strategies with less visibility.
The foundation-model market structure. Cursor’s adoption pattern (a large user base across many language ecosystems and many organizations) gives Anysphere meaningful market information about how AI-assisted coding actually works in production. The information advantage feeds back into Cursor’s product decisions and into the data that Cursor uses to train its in-house models. This is the kind of feedback loop that produces sustained competitive advantage and is the reason Cursor’s valuation is what it is.
The honest reading of Cursor in mid-2026 is that the company has executed exceptionally well on a product category that was still being defined when they started. The product is genuinely good, the strategic moves are sound, and the competitive position is strong but not unassailable. The next two years will determine whether Cursor consolidates its lead or whether competitors close the gap.
Frequently asked questions
Is Cursor based on Visual Studio Code? Yes, Cursor is a fork of VS Code. The codebase started from VS Code and has diverged through Cursor-specific changes, but most VS Code features and extensions work in Cursor without modification.
Do my VS Code extensions work in Cursor? Most of them, yes. Cursor uses a compatible extension system and the major language servers, formatters, debuggers, and theme packs work without modification. Some extensions that hook into VS Code’s specific update cycle or rely on Microsoft-specific telemetry may have issues.
Can I import my VS Code settings into Cursor? Yes. Cursor has a settings import option in the welcome flow that brings over keybindings, themes, snippets, and most configuration from a VS Code installation.
Which underlying models does Cursor use? Cursor supports Anthropic Claude (Opus, Sonnet, Haiku), OpenAI GPT (5, 5.5), Google Gemini, and Cursor’s own Composer models (1, 2, 2.5). The user can select which model to use for each AI surface or let Cursor pick the recommended default.
Can I use my own API keys? Yes, Cursor supports configuring your own API keys for Anthropic, OpenAI, and Google. When you use your own keys, the token costs are billed to your accounts at those providers rather than counting against Cursor’s plan quotas.
Is Cursor available for Windows, Mac, and Linux? Yes, all three platforms. The application is built on Electron (like VS Code) and the experience is consistent across platforms.
Does Cursor work offline? Partially. The IDE features work offline (editing, file management, language servers). The AI features require an internet connection because they call cloud-hosted models. Cursor has no on-device model option as of mid-2026.
Is there an enterprise version with on-premise model hosting? The Enterprise tier supports various enterprise deployment options, but full on-premise model hosting is not a standard offering. Specific arrangements are negotiated case-by-case for enterprise customers with on-premise requirements.
How does Cursor compare to GitHub Copilot in cost? Cursor Pro at $20/month is the same price as GitHub Copilot Individual. The value comparison is workload-dependent; Cursor’s deeper AI integration (Composer in particular) is meaningfully more capable than Copilot’s equivalent features, while Copilot’s distribution through GitHub and Microsoft ecosystems is more frictionless for users already in those ecosystems.
Is Cursor SOC 2 compliant? Yes, Cursor has SOC 2 Type II compliance and supports the additional compliance frameworks (HIPAA, GDPR data handling) that enterprise customers typically require. The specific compliance posture is detailed in Cursor’s trust center documentation.
Should I switch from VS Code to Cursor? If you regularly use AI-assisted coding and want it more integrated than a plugin can provide, yes. If your workflow is happy with current VS Code plus Copilot or another AI plugin, the marginal benefit of switching is smaller. The switching cost is low (your extensions and settings carry over) so the right experiment is to try Cursor for two weeks and see if the deeper AI integration meaningfully changes your workflow.
