Cursor / Anysphere

Bootstrapping Composer with autoinstall

Development environments for your cloud agents

Introducing Composer 2.5

A technical report on Composer 2

Improvements to Cursor Automations

You can now see a breakdown of your agent's context usage in Cursor 3.3. Use these stats to diagnose context issues and improve your setup across rules, skills, MCPs, and subagents.

We use previous generations of Composer to train future ones. Our autoinstall system has earlier Composer models set up dev environments for RL training. That way, the next generation can focus on learning to solve harder problems. https://cursor.com/blog/bootstrapping-composer-with-autoinstall

Introducing /orchestrate, a skill that recursively spawns agents to tackle your most ambitious tasks with the Cursor SDK. We’ve used it to: - Autoresearch our internal skills, cutting token use by 20% while improving evals - Cut cold start times on our internal backend by 80%

This builds on our existing research on multi-agent systems, with a key addition: verifiers. Planners spawn workers that write code and verifiers that run it. If verification fails, the planner spawns a new worker to fix it.

To try the /orchestrate skill with the Cursor SDK, download the plugin from our marketplace: https://cursor.com/marketplace/cursor/orchestrate

A new PR review experience is now available in Cursor 3. Take PRs from creation to merge, all in one place. You can see comments, diffs, commits, and review status to understand what changed and next steps. Navigate larger PRs more quickly with the file tree and changes picker.

Cursor can now split plans up into parallel subtasks by multitasking. Click "Build in Parallel" to have it identify independent tasks and run them simultaneously using async subagents.

When multitasking, Cursor can now break up your diffs into smaller, more mergeable PRs with the "Create PRs" quick action. It will identify logical slices and propose a split plan for your approval.

Pin your most commonly used skills as quick-action pills for easy access.

See everything new in Cursor: https://cursor.com/changelog/05-07-26

Composer 2 is 50% off in the SDK this weekend. Enjoy!

We’re introducing the Cursor SDK so you can build agents with the same runtime, harness, and models that power Cursor. Run agents from CI/CD pipelines, create automations for end-to-end workflows, or embed agents directly inside your products.

With the Cursor SDK, you can run agents locally or deploy them in our cloud.

We've open-sourced a few starter projects for you to build on: a coding agent CLI, a prototyping tool, and an agent-powered kanban board. Use Cursor to customize them for your use case: https://github.com/cursor/cookbook

Customers like Rippling, Notion, C3 AI, and Faire are using the Cursor SDK to build custom background agents, take bugs from ticket to merge-ready PR, and maintain self-healing codebases. Learn more: https://cursor.com/blog/typescript-sdk

Our agent harness makes models inside Cursor faster, smarter, and more token-efficient. Here's how we test improvements to the harness, monitor and repair degradations, and customize it for different models. https://cursor.com/blog/continually-improving-agent-harness

Cursor Security Review is now available for Teams and Enterprise plans. Run two types of always-on agents: 1. Security Reviewer checks every PR for vulnerabilities and leaves comments. 2. Vulnerability Scanner runs scheduled scans of your codebase and posts findings in Slack.

Customize these Cursor-managed security agents to match your team’s requirements. Adjust triggers, add your own instructions, give them custom tooling, and choose how outputs are shared.

We’re continuing to improve the runtime, harness, and models powering Cursor Security Review for a strong out-of-the-box experience. Security agents draw from your existing usage pool. Learn more: https://cursor.com/docs/security-review

GPT-5.5 Launches in Cursor, Tops CursorBench at 72.8% with 50% Discount via OpenAI Partnership

GPT-5.5 is now accessible in Cursor and leads CursorBench with a 72.8% score. Cursor has partnered with OpenAI to provide this model at 50% discount until May 2. CursorBench serves as the evaluation benchmark for model performance in Cursor.

GPT-5.5 Launches in Cursor, Tops CursorBench at 72.8% with 50% Discount via OpenAI Partnership

GPT-5.5 is now accessible in Cursor and leads CursorBench with a 72.8% score. Cursor has partnered with OpenAI to provide the model at 50% discount until May 2. CursorBench serves as the evaluation benchmark for model performance in Cursor.

Cursor 3.2 Enables Parallel Agent Execution via Multitasking, Worktrees, and Multi-Root Workspaces

Cursor 3.2 introduces /multitask for running async subagents to parallelize requests, bypassing queues by processing queued messages concurrently. Enhanced worktrees allow isolated background tasks across branches, with one-click integration to the local foreground. Multi-root workspaces support cross-repo changes in a single agent session, streamlining complex development workflows.

Cursor 3.2 Introduces Multitask Subagents and Worktrees for Parallelized AI Coding

Cursor 3.2 adds /multitask in the new interface, enabling async subagents to parallelize requests instead of queuing them sequentially. Queued messages can be multitasked without waiting for prior runs. Enhanced worktrees allow isolated background tasks across branches, multi-root workspaces support cross-repo changes in a single agent session, and branches can be pulled into the local foreground with one click.

Cursor 3.2 Enables Multi-Root Workspaces for Agent-Driven Cross-Repository Changes

Cursor 3.2 introduces multi-root workspaces, allowing a single agent session to target reusable workspaces composed of multiple folders for cross-repo changes. This builds on parallelization features like improved worktrees for isolated background tasks across branches and /multitask for async subagents. Users can download Cursor 3.2 to access these agents window enhancements.

Cursor 3.2 Introduces Multitasking Agents for Parallelized Development Workflows

Cursor 3.2 adds /multitask in the interface to run async subagents in parallel, bypassing queues for queued messages. Enhanced worktrees enable isolated background tasks across branches with one-click foreground integration. Multi-root workspaces support cross-repo agent sessions on combined folders.

AIE CODE Day 2: ft Google Deepmind, Anthropic, Cursor, Netflix, Cline, OpenAI, Meta, and METR

AIE Miami Day 2 ft. Cerebras, OpenCode, Cursor, Arize AI, and more!

Mention @​Cursor to kick off tasks in Slack and see updates of its work streaming in real time. Cursor uses context in the thread and broader channels to create a PR for you to review and ship.

Add Cursor to Slack: https://cursor.com/api/install-slack-app

We've reduced memory crashes in the Cursor desktop application by 80% since February. Here's how we detect, debug, and prevent OOMs at scale.

Read more here: https://cursor.com/blog/app-stability

We're partnering with SpaceX to improve Composer. https://cursor.com/blog/spacex-model-training

The shift to harder tasks wasn't immediate. Developers first used better models to do more work of similar complexity. Only after a 4-6 week lag did they start tackling more ambitious problems.

As AI gets better at writing code, the developer's job shifts toward managing that output. The fastest-growing task categories: documentation (+62%), architecture (+52%), code review (+51%), and learning (+50%). UI/styling grew just 15%.

Read the full research paper and blog post: https://cursor.com/blog/better-models-ambitious-work

Through the end of this weekend, we are doubling Composer 2 usage limits inside of Cursor's new agents window. Enjoy!

Download Cursor to use the agents window: https://cursor.com/download

We recently shipped quality-of-life improvements to the Cursor CLI to make working with agents in the terminal more delightful. Use /debug to find root causes and fix tricky bugs that are hard to reproduce or understand.

/btw allows you to ask a quick side question without derailing the agent's current run.

/config opens a settings panel inside the CLI where you can view and change default behaviors, runtime preferences, and more. You can also ask Cursor to apply configuration changes for you using the skill /update-cli-config.

Customize the CLI status bar using /statusline to surface the session and runtime signals you care about.

Download the Cursor CLI: https://cursor.com/cli

The Future of Coding Agents with Sasha Rush (Cursor/Cornell)

Developers Shift Coding Agents from Short Tasks to Long-Running Autonomous Cloud Systems

Developers are transitioning coding agents from brief 5-15 minute local runs to persistent cloud-based agents operating asynchronously for hours or days within VPCs, enabling completion of full bugs, features, or projects. This involves self-hosting private workers for extended autonomy. Executives, lacking direct experience, test AI tools on side projects and observe substantial productivity gains over traditional methods.

AI Coding Agents Signal Imminent Action-Oriented Shifts Across Professional Domains

Cursor/Anysphere's AI coding agents are currently limited by running on only 68 15-minute intervals, creating a performance bottleneck with significant room for improvement. Coding advancements in 2025 serve as the bellwether for AI's evolution from query response to autonomous action-taking. This shift will soon extend to finance, recruiting, legal, sales, and other functions, transforming professional workflows.