Cursor 3.2 ships /multitask — spawns parallel subagents instead of serializing them, the first single-IDE answer to Antigravity's multi-agent bet
Cursor 3.2 added the /multitask command that spawns parallel subagents inside the IDE rather than serializing them. The feature is the first single-IDE response to Antigravity 2.0's multi-agent orchestration positioning, and it lands the same week as Cursor Composer 2.5. Cursor is staking out the position that parallel agents inside a familiar editor surface is the right developer pattern, not a separate runtime.
The product-positioning subtext is the interesting part. Antigravity 2.0 made the bet that multi-agent orchestration deserved a new runtime with its own surface (built-in Chromium browser, hosted SDK, scheduled tasks). Cursor's /multitask makes the opposite bet: developers want parallel agents but they want them inside the editor they already live in, not in a new product surface they have to learn.
The Cursor Composer 2.5 release alongside /multitask is the broader feature set: improved code-context retrieval, longer agent action chains, better diff review surfaces. The cumulative effect is that Cursor is now feature-comparable with Antigravity 2.0 on the multi-agent dimension without sacrificing the IDE-native UX advantage. The market answer to which pattern wins — Cursor's in-IDE or Antigravity's separate runtime — will probably bifurcate by developer type, with senior developers preferring the in-IDE pattern and operator-developers preferring the runtime pattern.
Lushbinary — AI Coding Agents 2026: Claude Code vs Antigravity 2.0 vs Codex vs Cursor → · Amit Ray — Best AI Coding Tools 2026: Antigravity vs Cursor vs Windsurf vs v0 vs Copilot → · Stormap — Google Antigravity vs Cursor vs GitHub Copilot: Which AI IDE Wins in 2026? →