last_validated: 2026-04-06 decay_rate: fast
Cursor's $2 billion bet: The IDE is now a fallback, not the default
Source: https://thenewstack.io/cursors-2-billion-bet-the-ide-is-now-a-fallback-not-the-default/ Date: April 5, 2026 Author: Janakiram MSV Publication: The New Stack
Summary
Cursor 3 (codename "Glass") represents a fundamental architectural shift in developer tooling: the company rebuilt its product from scratch with an agent management console as the primary interface, demoting the traditional IDE to a secondary fallback surface. The prompt box now sits where the file tree used to be, signaling the product's assumption that engineers will spend most of their time dispatching agents, reviewing output, and deciding what ships rather than writing code directly. The release came during an intense six-week product offensive responding to Claude Code's $2.5B run rate overtaking Cursor's $2B, which included three other major launches: Automations (GitHub/Slack-triggered agents), Composer 2 (Cursor's first in-house model built on Kimi K2.5 claiming to beat Opus 4.6 at lower cost), and self-hosted cloud agents for Fortune 500s. Cursor 3's standout feature is Cloud Handoff—seamless session portability letting developers move running agent sessions between laptop and cloud mid-task. The article frames this as part of a broader architectural divergence in developer tooling: Anthropic went terminal-first (CLI as orchestration layer, no IDE), OpenAI went omni-surface (orchestration everywhere via desktop/CLI/web/extensions), Google went dual-mode (Antigravity's separate Editor View and Manager Surface), while Cursor went agent-first-with-IDE-fallback (agent console default, editor secondary). The pattern mirrors infrastructure's evolution from SSH terminals to cloud dashboards, but with higher stakes because the abstraction being demoted—the code editor—is what developers have used for 40 years.
Key Claims
- Cursor 3 (codename "Glass") was "built from scratch, centered around agents" as a sibling to the original VS Code fork, not an evolution of it
- The agent management console is the primary interface with the traditional IDE available as a secondary surface; "the prompt box sits where the file tree used to be"
- Cursor's annualized revenue crossed $2 billion in February 2026, doubling in three months (Bloomberg); Claude Code reached $2.5 billion run rate with 300,000+ business customers (Fortune)
- Developers were publicly posting about leaving Cursor for Claude Code, and one Cursor investor told Fortune that several portfolio startups were "decoupling from it"
- Cursor launched four major products in six weeks (March 5 - April 5, 2026): Automations (GitHub/Slack/timer-triggered agents), Composer 2 (March 19), self-hosted cloud agents, and Cursor 3
- Composer 2 is Cursor's first in-house coding model, built on Kimi K2.5 with "75% of compute from Cursor's own RL post-training on usage data"; scored 61.3 on CursorBench vs. Opus 4.6 at 58.2 (caveat: CursorBench is Cursor's proprietary eval)
- Cursor 3's workspace is multi-repo by default with unified sidebar showing all local and cloud agents from every surface (mobile, web, Slack, GitHub, Linear); cloud agents generate demos and screenshots of their work
- Cloud Handoff allows moving running agent sessions between laptop and cloud mid-task, letting work continue while machine is closed; "that kind of session portability between local and cloud environments has been a gap in most competing tools"
- Cursor acquired code review platform Graphite in December 2025 because "reviewing code was becoming the bottleneck as AI accelerated writing it" (CEO Michael Truell)
- The architectural divergence in developer tooling: Anthropic (terminal-first, no IDE), OpenAI (omni-surface orchestration via desktop/CLI/web/extensions), Google Antigravity (dual-mode with coequal Editor View and Manager Surface), Cursor (agent-first with IDE fallback as default)
- Google paid $2.4 billion in licensing fees to Windsurf and hired its CEO and engineers into DeepMind to build Antigravity
- Cursor's Composer 2 pricing: $0.50/M input tokens, $2.50/M output tokens (March 2026), "substantially below frontier model pricing"
- "Engineers using Cursor 3 spend their time reviewing diffs generated by agents, verifying screenshots of what cloud agents produced, deciding which tasks to push to the cloud and which to keep local, and managing PR workflows. That's a different skill set than writing code."
- The role of software engineer is "converging with that of a systems operator who works at the application layer"
- "If the agent-first interface wins, VS Code extensions become less relevant" — Microsoft "should be paying close attention" as "the assumption that VS Code is the center of gravity for developer tooling, an assumption that has held for nearly a decade, is weakening"
- Model choice is now an infrastructure decision like choosing a database or cloud region; token economics compound at scale for teams running dozens of parallel agents
- "The code editor defined how software got built for four decades. Cursor 3 is a bet that supervising agents will matter more than editing files."
Tags
#cursor #cursor-3 #glass #agent-management-console #cloud-handoff #composer-2 #kimi-k2.5 #anthropic #claude-code #openai-codex #google-antigravity #windsurf #graphite #vscode #developer-tooling #ai-coding-agents #orchestration-layer #session-portability #multi-repo-workspace #token-economics #model-choice
Related
- ai agent ecosystem — agent management console as primary interface, Cloud Handoff, orchestration layer architecture
- vertical models and usage data — Composer 2 as vertical coding model (Kimi K2.5 + RL on Cursor usage data)
- open models and local inference — Kimi K2.5 as base model for Composer 2
- token economics and pricing — model choice as infrastructure decision, token economics driving tool selection at scale
- ainews every lab serious enough about — developer toolchain land grab context (GDM/Antigravity, Anthropic/Bun, OpenAI/Astral)
- ai daily brief anthropic mythos vertical models — Composer 2's performance claims (beats Opus 4.6) connect to vertical models thesis