The latest AI news and updates of Cursor — the AI-first code editor built for pair-programming with AI. Covering Cursor's latest product updates, research, and launches from the past 90 days.
Cursor15h agoCursor now enables Auto-review by default for all new users. This feature uses a classifier subagent to evaluate agent actions in context, deciding whether to allow, block, or request approval. Internal evaluations show the classifier achieves 97% accuracy, helping to minimize approval fatigue while maintaining safety during autonomous coding tasks.
Cursor16h agoCursor updated its Bugbot code review agent to be 3x faster, 22% cheaper, and 10% more effective at finding bugs. The release introduces a /review command for local pre-push checks and an incremental review mode that only scans new code changes. These performance gains are powered by the new Composer 2.5 model, available now in Cursor 3.7.
CursorJun 10Cursor, an AI-first code editor, has made Anthropic's Claude Fable 5 model available within its platform. The model achieved a new state of the art on CursorBench 3.1 with a score of 72.9%, surpassing the previous best by 8 points. This update signifies a notable improvement in AI coding capabilities for complex development tasks.
CursorJun 5Cursor has updated its Design Mode, enabling users to visually guide AI agents by pointing, drawing, or speaking directly on a running application's interface. This update aims to streamline UI development by providing agents with precise visual context for code changes.
CursorJun 4Cursor has updated its canvases feature, allowing users to publish and share AI-generated interactive applications like dashboards and internal tools via URL. This release also introduces a Design Mode for direct UI editing and an interactive report to visualize agent context usage. The enhancements aim to make AI-driven application development more collaborative, transparent, and iterative for teams.
CursorJun 1Cursor is restructuring its Teams plan with increased usage limits and a new high-capacity seat tier for power users. The update introduces separate usage pools for first-party and third-party models to improve cost predictability as autonomous coding workflows scale.
CursorMay 29Cursor launched Auto-review, a new run mode that delegates tool call approvals to a specialized classifier subagent. By automating safety checks for terminal and API actions, the system allows coding agents to work longer without human interruption.
CursorMay 28Cursor added Anthropic's Claude Opus 4.8 to its AI-native code editor, delivering a performance jump on internal benchmarks for complex, multi-step tasks. The update introduces a thinking-token variant and a significantly cheaper fast mode, prioritizing model persistence across large codebases.
CursorMay 22Cursor released a Python SDK in public beta that allows developers to programmatically invoke the same agentic runtime and models found in its code editor. This expansion brings Cursor's orchestration layer to the Python ecosystem, enabling automated repository management and custom CI/CD agents.
CursorMay 20Cursor 3.5 integrates always-on automations directly into the Agents Window, allowing users to manage background tasks and active agent sessions in a single workspace. The update introduces repo-less automations that can perform general engineering tasks, like Slack thread prioritization, without requiring access to a specific codebase.
CursorMay 19Cursor launched a Jira integration that allows teams to trigger autonomous cloud agents directly from project management tickets. By assigning a work item or mentioning the agent in a comment, developers can offload bug fixes and feature additions that result in merge-ready pull requests.
CursorMay 18Cursor released Composer 2.5, a coding model optimized for sustained performance on complex, multi-step engineering tasks. The update introduces a new reinforcement learning method that provides localized feedback during long trajectories to reduce errors in tool use and communication.
CursorMay 14Cursor introduced new tools to configure and govern cloud development environments, allowing AI agents to operate across multiple repositories with full access to build systems and credentials. This shift provides agents with the same infrastructure as human engineers, enabling them to verify their own code and close the loop on complex tasks.
CursorMay 11Cursor introduced customizable effort levels for Bugbot, allowing developers to choose how deeply the AI agent analyzes pull requests. By scaling inference-time compute, the high-effort mode detects 35% more bugs than the default setting while maintaining high resolution rates.
CursorMay 7Cursor launched the /orchestrate skill for its SDK, allowing AI agents to recursively spawn sub-agents to handle complex engineering tasks. This moves agentic coding into a self-correcting loop where planners spawn workers to write code and verifiers to validate the output.
CursorMay 7Cursor 3 now features a native pull request review experience that allows developers to manage comments, diffs, and merges without leaving the IDE. This shift consolidates the entire development cycle into a single interface, reducing the context switching that typically stalls AI-assisted workflows.
CursorMay 7Cursor 3.3 added a context usage breakdown that shows exactly how AI agents consume their token budget across rules, skills, and external tools. This shifts agentic coding from a black-box experience to a transparent one where developers can profile and optimize the information fed to the model. By identifying which rules or subagents are crowding the context window, users can reduce hallucinations.
CursorMay 7Cursor revealed its autoinstall system, which uses previous generations of its Composer model to automatically set up and verify runnable development environments for reinforcement learning training. By using AI to handle the plumbing of environment configuration and dependency mocking, the team ensures that student models receive a clean reward signal instead of wasting compute debugging setup failures.
CursorApr 30Cursor shared a technical deep dive into its agent harness, the orchestration layer that manages context, tools, and error correction for its AI coding agents. The update reveals that agent performance depends less on raw model power and more on specialized engineering like dynamic context discovery and model-specific tool tuning. This shift highlights why the harness is becoming the primary competitive moat for AI-native development tools.
CursorApr 30Cursor launched Security Review for Teams and Enterprise plans, featuring managed agents that audit pull requests and perform scheduled codebase scans. This shifts security from a manual checkpoint to an autonomous, always-on background process that scales with AI-driven development. The system uses LLMs to verify if reported vulnerabilities are actually resolved by analyzing incremental code diffs.
CursorApr 30Cursor released a TypeScript SDK in public beta that allows developers to programmatically invoke the same AI agents and orchestration layer found in its code editor. This shifts agentic coding from an interactive IDE feature to a programmable infrastructure layer that can be embedded into CI/CD pipelines and third-party products.
CursorApr 24Cursor integrated OpenAI's GPT-5.5, which now leads its internal CursorBench evaluation with a 72.8% score. The update highlights a shift toward using real-world developer sessions rather than public benchmarks to measure frontier model performance on complex, multi-file coding tasks.
CursorApr 24Cursor 3.2 introduces the multitask command to run asynchronous subagents in parallel rather than waiting for sequential queue execution. This update shifts the developer experience from serial task management to simultaneous orchestration across multiple repositories within a single agent session.
CursorApr 24Cursor launched a Slack integration that allows teams to initiate coding tasks and generate pull requests by mentioning the AI agent in a conversation. This moves agentic coding from the IDE into the team's communication layer, allowing agents to use chat context to autonomously ship code.
CursorApr 24Cursor is partnering with SpaceX to utilize xAI's Colossus infrastructure for training its Composer agentic coding models. By removing compute bottlenecks, the team aims to significantly increase model intelligence through scaled reinforcement learning and continued pretraining.
CursorApr 21Cursor reduced desktop application memory crashes by 80% since February by implementing a dual-strategy debugging system that monitors memory pressure and automates fixes. As AI agents take on more complex tasks like web browsing and codebase indexing, the IDE must evolve from a text editor into a high-performance runtime.
CursorApr 18Regex searches in large codebases can take over 15 seconds, stalling AI agents mid-task. Cursor's Instant Grep indexes files locally, cutting latency to 13 milliseconds. Agents in big monorepos can now iterate without multi-second delays.
CursorApr 16A study of 500 companies by Cursor and the University of Chicago found that as AI models improve, developers tackle 68% more high-complexity tasks. Rather than working less, teams are using efficiency gains to expand their project scope and shift their focus toward system architecture and code review.
CursorApr 15Cursor 3.1 introduced Canvases, a feature that allows AI agents to generate interactive interfaces like dashboards and diagrams instead of plain text. This shift increases information bandwidth by letting users explore non-linear data visualizations for tasks like PR reviews and system monitoring.
CursorApr 15Cursor added event-based triggers for Sentry, allowing AI agents to autonomously investigate production errors and open pull requests with fixes. This shift moves AI from a reactive co-pilot to an active on-call engineer that remediates bugs the moment they are detected.
CursorApr 15Cursor partnered with NVIDIA to apply a multi-agent system to CUDA kernel optimization, achieving a 38 percent geomean speedup on Blackwell GPUs. This demonstrates that autonomous agents can solve complex hardware engineering tasks that previously required months of manual effort from human experts.
CursorApr 9Cursor's autonomous cloud agents can now automatically include visual proof of their work directly in GitHub Pull Requests. This provides immediate verification for human reviewers, moving AI from a black-box code generator to a transparent teammate that demonstrates its results.
CursorApr 9Cursor's code review agent, Bugbot, can now automatically generate repository-specific rules by analyzing developer reactions and human reviewer comments on pull requests. This real-time learning loop has increased the agent's issue resolution rate to 78%, significantly reducing false positives.
CursorApr 7Anysphere rebuilt the Mixture of Experts inference path for NVIDIA Blackwell GPUs, achieving 1.84x faster throughput by assigning GPU warps to individual output neurons. This warp decode approach eliminates the data-shuffling overhead typical of expert-centric models while improving output accuracy by 1.4x.
CursorApr 2Cursor 3 introduces a dedicated interface for orchestrating multiple autonomous agents across different repositories simultaneously. This shift moves developers from manual file editing to a high-level management role where they verify and hand off tasks between local and cloud environments.
CursorMar 28Cursor is now using real-time reinforcement learning to train its Composer agent using actual user interactions as reward signals instead of relying solely on simulations. This pipeline allows the team to deploy an improved model checkpoint every five hours, ensuring the training data remains on-policy. By using real-world human oversight as the ground truth, the system eliminates the modeling errors typically found in simulated environments.
CursorMar 26Cursor can now create and modify Figma components directly on the canvas via Figma's new use_figma MCP tool, now in open beta. Coding agents work within your team's actual design system, guided by markdown skills that encode your conventions.
CursorMar 26Cursor published a technical report on Composer 2, a coding agent trained via pretraining on Kimi K2.5 and RL on real engineering tasks. It scores 61.3 on CursorBench — 37% above Composer 1.5 — matching frontier models at lower cost.
CursorMar 25Cursor has made self-hosted cloud agents generally available, letting enterprise teams run autonomous coding agents entirely within their own network. Regulated organizations that previously couldn't use cloud agents due to compliance rules now have a path to adopt them.
CursorMar 20Cursor released Composer 2, its latest coding model, now available in Cursor. It scores 61.3 on CursorBench — up from 44.2 on Composer 1.5 — and is priced at $0.50/M input and $2.50/M output tokens.
CursorMar 20Cursor trained Composer, its agentic coding model, to self-summarize through reinforcement learning rather than prompt-based compaction. This cuts context compaction error by 50% while using one-fifth the tokens, letting Composer handle tasks requiring hundreds of turns.
CursorMar 18Cursor built a fleet of security agents using Cursor Automations that reviews 3,000+ internal PRs weekly, catching 200+ vulnerabilities. Now it's releasing four automation templates so any team can replicate the same security agent architecture on their own codebase.
Frequently asked questions