Is VS Code Obsolete? Cursor 3 Redefines the IDE Around AI Agents

What Cursor 3 Brings

Cursor 3 (code‑named Glass) is built from scratch with an intelligent‑agent management console as the primary interface, relegating the classic IDE to a secondary view. Engineers can still write code, but the design assumes most of their time will be spent scheduling agents, reviewing output, and deciding which tasks to publish. The file‑tree view is replaced by a prompt‑input box.

The workspace supports multiple repositories by default, showing both local and cloud agents in a unified sidebar that aggregates agents from desktop, web, Slack, GitHub, and Linear. Cloud agents generate screenshots of their work, letting engineers inspect changes without pulling code locally.

The standout feature is Cloud Handoff , which lets a running agent session move from a laptop to the cloud and back, ensuring work continues when the machine is off. This capability addresses a common shortcoming in competing products.

Pressure Behind the Transformation

Cursor’s rapid product push follows a “product acceleration” campaign triggered by market pressure: a Fortune article labeled the company as a classic “innovator’s dilemma,” and Bloomberg reported its annual revenue surpassing $2 billion in February 2026, doubling in three months.

In response, Cursor launched three initiatives in March: the Automations system (triggering agents from GitHub events, Slack messages, or timers), Composer 2 (built on the open‑source Kimi K2.5 model), and self‑hosted cloud agents for Fortune‑500 customers.

Composer 2 scored 61.3 on Cursor’s proprietary CursorBench, beating Claude Opus 4.6’s 58.2, with lower token costs. Pricing (as of March 2026) is $0.50 per M input tokens and $2.50 per M output tokens, notably cheaper than Anthropic and OpenAI models.

Structural Shift in IDE Design

All major AI‑assisted development players agree that agents need a dedicated orchestration interface, but they differ on placement. Anthropic’s Claude Code adopts a terminal‑first approach, treating the CLI as the orchestration layer. OpenAI’s Codex offers a cloud UI, desktop app, and IDE extensions, viewing the orchestration layer as omnipresent. Google’s Antigravity provides both an editor view and a manager view within the same app.

Cursor follows Google’s integrated model but makes the agent console the default view, whereas Google treats both views equally. This reflects differing judgments about where developers will spend most of their time.

“For the past forty years, software development paradigms have been defined by code editors. Cursor 3 bets that supervising agents will outweigh editing files.”

Implications for Developers

Model choice becomes infrastructure: Cursor sets Composer 2 as the default model but allows switching to Claude, GPT‑5.4, or Gemini. Selecting a model is now akin to choosing a database or cloud region, with token economics scaling with usage.

VS Code’s moat is eroding: Cursor, a VS Code fork, differentiates by centering agents. If agent‑centric interfaces win, VS Code extensions lose relevance, and Microsoft must monitor this trend.

Workflow and role changes: Engineers will spend significant time reviewing agent‑generated diffs, validating cloud screenshots, and managing PR flows, shifting the skill set toward platform‑ops and engineering‑management rather than pure coding.

“Software engineers’ roles are merging with application‑layer system operations.”

Future Outlook

The orchestration layer for AI coding agents is emerging as a new product category. The open question is whether this layer should live inside IDEs, outside them, or both. Anthropic and OpenAI favor independent tools, while Cursor and Google back integrated consoles. The winner may shape developer tooling for the next decade, much like the early cloud control‑plane battle defined today’s infrastructure giants.