Cursor 3 Launch: IDEs Become Secondary as AI Agent Consoles Take Over

Cursor 3 has replaced traditional code editors with an AI agent management console, signaling a major shift in AI-assisted development tools and developer workflows.

As the fastest-growing AI code editor in terms of revenue, Cursor has launched its first non-code editor product. Cursor 3 (codenamed Glass) is built from the ground up, positioning the agent management console as the main interface, relegating traditional IDEs to a secondary role. Engineers can still write code within it, but the core design philosophy has shifted to “users will spend most of their time scheduling agents, reviewing outputs, and deciding which tasks to push to the cloud.” The area previously displaying the file tree has now been replaced by a prompt input box.

In recent months, the status of coding tools has been increasingly overshadowed by AI models. The emergence of Cursor 3 is a direct result of this $2 billion revenue company embracing this reality and betting its entire product line on it. This transformation mirrors the past experiences of infrastructure engineers.

Cloud control panels have replaced SSH terminals, and Kubernetes controllers have taken over manual server configurations. Now, orchestration layers are replacing code editors as the primary interface. Although the trend is similar, the risks of this wave of change are more pronounced, as the downgraded abstraction layer is the foundational architecture used by developers for the past 40 years.

What Cursor 3 Brings

Cursor was born in 2022 as a fork of VS Code, and it has now evolved into a fully agent-centric version. Cursor claims that the new interface is “completely built from scratch with agents at the center,” viewing traditional IDEs as an option to switch to at any time.

The workspace supports multiple repositories by default, allowing both agents and users to operate in different repositories simultaneously, with all local and cloud agents displayed in a unified sidebar. This sidebar extracts agents from all Cursor interfaces, including sessions initiated from mobile devices, web clients, Slack, GitHub, and Linear. Cloud agents can generate demonstrations and screenshots, enabling engineers to quickly understand changes without pulling code locally.

The standout feature of the new version is Cloud Handoff. Users can transfer running agent sessions from their laptops to Cursor’s cloud, ensuring they continue to operate when the computer is shut down, and can be pulled back locally for editing and testing when ready. Users can initiate tasks in the cloud and retract them when they need to take control. This ability to migrate sessions between local and cloud environments has been a significant shortcoming for most competing products.

The launch of Cursor 3 can be seen as a major shift from managing a single server to managing server clusters through a control plane. While users can still access servers via SSH when necessary, the control plane has become the hub for decision-making, workload distribution, and system status viewing. Cursor 3’s approach to agents also marks a significant transformation, where the IDE is akin to SSH, and Glass represents the control plane.

Transformation Driven by Pressure

The release of Cursor 3 is not coincidental. Over the past six weeks, Cursor has undertaken what I call an “accelerated product offensive.” In late March, Fortune magazine published a feature describing Cursor’s situation as a classic case of the “innovator’s dilemma,” which quickly sparked strong reactions. Previously, Bloomberg had pointed out that Cursor’s annualized revenue officially surpassed $2 billion in February 2026, doubling within three months.

However, Fortune discovered that Anthropic’s terminal-first programming agent, Claude Code, had raised its annualized revenue to $2.5 billion in just over a year, winning over 300,000 enterprise customers. Developers have publicly expressed their intent to leave Cursor for Claude Code. One Cursor investor told Fortune that several startups in his portfolio are considering terminating their agreements with Cursor. Reports in March indicated that despite Cursor seeking a new round of financing with a valuation of about $50 billion, market confidence in its future trajectory was eroding.

In response, Cursor quickly implemented three major initiatives to tackle these challenges. On March 5, the company launched the Automations system, which triggers agents based on GitHub events, Slack messages, and timers without human intervention.

On March 19, Cursor released Composer 2, its first self-developed model based on the open-source Kimi K2.5 from Moonlight.

Cursor claims that Composer 2 scored 61.3 in its proprietary CursorBench tests, surpassing Claude Opus 4.6’s score of 58.2, with lower per-token costs. It is important to note that CursorBench is an evaluation suite developed by Cursor itself.

Also in March, Cursor enabled self-hosted cloud agents, allowing Fortune 500 companies to run Cursor agents on their internal infrastructures.

Then came the official release of Cursor 3. Three products were launched within a month, and the interface underwent a complete overhaul. This pace is a clear indication that the company believes its business field is undergoing a redefinition.

Structural Change

The significance of Cursor’s transformation lies in highlighting a paradigm shift that transcends a single company. Today, all major players in the AI-assisted development field have reached a consensus that agents need their own orchestration interface. The divergence among vendors lies in where this interface should reside, creating interesting differences in the architecture of developer tools.

Anthropic’s Claude Code adopts a terminal-first model, completely abandoning integrated development environments (IDEs). The command-line interface (CLI) serves as the orchestration layer, where developers work in the shell using natural language commands. Anthropic later added a browser-based interface and desktop applications at claude.ai/code, but still views the terminal as core, allowing the orchestration layer to operate independently of the editor.

OpenAI has chosen a different path. Codex now encompasses standalone desktop applications, command-line interfaces, IDE extensions for VS Code and its forks, along with a cloud interface at chatgpt.com/codex. The desktop application serves as a command center for managing parallel agents, viewing differences, and running work across projects. OpenAI’s philosophy is that the orchestration layer should be ubiquitous, allowing developers to access it regardless of the interface used.

“For the past forty years, software development paradigms have been defined by code editors. Cursor 3 bets everything on the belief that supervising agents will become more important than editing files.”

Google’s solution is closest to Cursor’s. After paying $2.4 billion for the Windsurf license and bringing its CEO and core engineers into DeepMind, Google launched Antigravity—a center-focused IDE that offers two distinct modes. The editor view includes a traditional programming environment, while the manager interface allows developers to generate, orchestrate, and observe multiple agents in parallel. Agent orchestration and code editing coexist within the same application but are presented by different views.

Cursor 3 adopts this internal IDE model but enhances it further. Glass sets the agent console as the default view, while the editor becomes an alternative view. Google Antigravity considers both views equally important, reflecting differences in emphasis. In other words, the two companies have differing judgments about where developers will ultimately spend their time.

We can draw an analogy between the evolution of cloud infrastructure and this divergence. When AWS launched its management console, users did not stop using SSH. However, SSH was no longer the place for making infrastructure decisions; the console began to serve as the control plane, while SSH became an occasionally used debugging tool.

Anthropic and OpenAI believe the orchestration layer can operate as a standalone application, separate from the IDE. Cursor and Google believe the orchestration layer needs to coexist with the editor, allowing developers to view code directly when agents make mistakes. The industry generally agrees that agent orchestration will become the new main interface, but consensus on specific architectural design has yet to be reached.

What This Means for Developers

Currently, the actual impact of this transformation is primarily reflected in three aspects:

1. Model Selection Becomes Infrastructure
   Cursor sets Composer 2 as the default model but still allows developers to switch to Claude, GPT-5.4, and Gemini based on the dialogue context. The model driving the agents has now become a decision at the infrastructure level, akin to choosing a database or cloud region, with token economic effects gradually forming as scale increases.

   Cursor’s standard version of Composer 2 is priced at $0.50 per million input tokens and $2.50 per million output tokens (as of March 2026), significantly lower than the pricing of Anthropic and OpenAI’s leading models. For teams running dozens of parallel agents, such pricing is undoubtedly more appealing than feature settings or UI preferences.

2. The Moat of VS Code is Drying Up
   Cursor forked from VS Code, inheriting its extension ecosystem while achieving instant distribution. Cursor 3 takes a different path, aiming to create a differentiated advantage. If an agent-centric interface design ultimately prevails, the importance of VS Code extensions will diminish. Microsoft clearly needs to closely monitor this trend.

   The assumption of VS Code as the center of developer tools has persisted for nearly a decade but is now gradually unraveling.

   JetBrains faces similar pressure. As the main interaction interface shifts from editing files to managing agents, traditional IDEs will also lose their competitive advantages in programming intelligence and refactoring tools.

3. Workflow Changes First, Job Roles Follow
   Cursor 3 believes that engineers will spend significant time reviewing differences generated by agents, validating screenshots generated by cloud agents, deciding which tasks to push to the cloud and which to keep local, and managing PR workflows.

   This is fundamentally different from previous coding methods and skill requirements, indicating that programming is shifting more towards engineering management or platform operations roles rather than traditional software developers. In other words, the role of software engineers is merging with that of application-layer system operators.

“The role of software engineers is merging with that of application-layer system operators.”

Cursor’s own development trajectory reflects this shift. The company acquired the code review platform Graphite in December 2025. As CEO Michael Truell stated, with AI accelerating code generation, code review is becoming the new bottleneck.

Cursor 3 extends this logic further, where agents write code, Graphite handles reviews, and engineers coordinate between the two. The importance of integrated development environments (IDEs) has thus been relegated to a secondary position.

Future Outlook

The orchestration layer for AI programming agents has now become a new product category. All major vendors have launched related products, and the unresolved question lies in the architectural level. Should the orchestration layer exist within the IDE, outside of it, or simultaneously across all interfaces?

Anthropic and OpenAI are betting on independent tool forms, while Cursor and Google are betting on integrated comprehensive consoles with IDEs. Whichever choice aligns more closely with the true trend will win the support of developers in the next decade, just as the cloud control plane battle in the early 2010s determined the current positions of infrastructure giants.

For the past forty years, the way software has been built has been defined by code editors. Cursor 3 bets on the idea that supervising agents is more important than editing files. Cursor does not aim to eliminate IDEs but rather to lower their status.

If this judgment proves accurate, what lies ahead may very well be the last code editor in human memory. The future remains to be seen.