Cursor Projects and cloud agents expand, Claude Code and Codex CLI add voice plus Bedrock (agent tools this week)

What shipped

Across this week Cursor, Claude Code, and OpenAI Codex released updates aimed at agentic coding workflows. Cursor emphasized multi-agent coordination and infrastructure controls. Claude Code and Codex CLI added terminal and integration improvements for daily use.

Cursor

Cursor shipped ten updates centered on cloud agents and the new Projects feature. These changes support long-running tasks with event triggers and isolated execution. Teams gain options to keep data on self-hosted machines while coordinating subagents across local and cloud setups.

• •Cursor Event Subscriptions Cursor cloud agents now subscribe to PR and Slack events, run subagents on isolated VMs, and accept mid-session steering for sustained goals like migrations.
• •Cursor Scratch Projects Cloud agents can start new projects without Git providers, create origin repos, forward browser previews, and deploy straight to Vercel.
• •Cursor Self-Hosted Machines Cursor added self-hosted machine support so code and secrets stay inside internal networks, plus dynamic pool scheduling and Linux or Mac computer use.
• •Cursor Projects Feature The Projects feature coordinates subagents for long tasks such as full apps, keeps shared context across machines, and reacts to external triggers.
• •Cursor Rollouts Bots Rollouts and Security Review bots for Teams plans monitor deployments and flag exploitable bugs in pull requests automatically.
• •Cursor Origin Hosting Origin native hosting in beta gives paid users built-in repos, GitHub sync, in-repo PR reviews, and extensions for Vercel and Buildkite.
• •Cursor Agent Enhancements Cloud agents gained PR and Slack subscriptions, pinned custom modes, isolated subagents, and the /goal command for long objectives.
• •Cursor No-SCM Projects Cloud agents now scaffold projects from prompts alone, save to Origin repos, and support live previews with Vercel deployment.
• •Cursor Worker Pools Self-hosted machines keep execution inside company networks, with dynamic scheduling, idle hibernation, and sandbox support on Linux and Mac.
• •Cursor Long-Running Workflows Projects in beta use coordinator agents to delegate work to parallel subagents and maintain context across machines and Slack triggers.

Claude Code

Claude Code released eleven updates focused on terminal stability, VS Code features, and gateway controls. Changes include better session handling, plugin security, and environment management for Team plans. The updates target both local TUI users and cloud gateway setups.

• •Claude Code Bedrock Support Claude apps gateway added Bedrock assume_role and guardrail options plus MCP URL-mode elicitation for browser flows.
• •Claude Code Prose Width A new maxProseWidth setting caps terminal output, while session resume fixes improve extended thinking retention and vim mode.
• •Claude Code VS Code Memory The VS Code extension now lets users view, edit, and delete saved memories and send attached images without extra text.
• •Claude Code NPM Security Plugins installed via npm now use npm pack with ignore-scripts and integrity checks to block automatic script runs.
• •Claude Code Skill Sync Skills and plugins from claude.ai now sync automatically to terminal sessions, and Ctrl+Enter sends queued messages right away.
• •Claude Code Proxy Fix A 400 error tied to advisor_20260301 was fixed for users routing requests through proxies or gateways.
• •Claude Code Cloud Views Team and Enterprise plans gained Personal and Organization environment views on the web for sharing and session management.
• •Claude Code VS Code Controls Sign out, tasks panel, token usage metrics, and per-change diff reviews were added to the VS Code extension.
• •Claude Code TaskOutput Removal The TaskOutput tool was dropped in favor of direct Read tool access to background files, with prompt sanitization added.
• •Claude Code AGENTS.md Fallback Claude Code now reads AGENTS.md when CLAUDE.md is absent and supports forward proxy egress on gateways.
• •Claude Code Auto Mode Default Auto mode switched to a server-side classifier for most users, removing local classifier charges and adding status rows.

OpenAI Codex

OpenAI Codex CLI shipped eleven updates including new GPT-6 models, voice mode, and Bedrock credential handling. Terminal improvements cover transcripts, daemon recovery, and task management. These changes target both local TUI sessions and cloud provider integrations.

• •Codex CLI GPT-6 Models Codex CLI 0.157.0 added GPT-6 Sol and Luna support with Amazon Bedrock integration and fullscreen transcripts by default.
• •Codex CLI Voice Mode Experimental voice conversations with live transcripts arrived in the TUI along with Touch ID verification for MCP requests on macOS.
• •Codex CLI Reasoning Fix Version 0.155.1 disables TUI reasoning summaries by default to avoid rejections from unsupported providers.
• •Codex iOS Folders ChatGPT for iOS gained folder creation from the file picker and writing blocks with draft alternatives.
• •Codex CLI Streaming Summaries The TUI now shows live reasoning summaries in the status row and records turn completion timestamps.
• •Codex Bedrock Credentials Amazon Bedrock integrations can fetch AWS credentials via commands with caching and expiration refresh built in.
• •Codex Daemon Scheduling Configurable daemon update schedules and explicit CLI commands let saved threads recover after restarts.
• •Codex Agent Overview Users can hide, archive, or delete tasks and view worktree ownership details in the agents overview.
• •Codex Touch ID MCP Local TUI sessions on supported Macs use Touch ID and Secure Enclave signing for MCP requests.
• •Codex Experimental Voice The /voice command adds live transcripts and animated split-flap tiles for microphone-controlled conversations.
• •Codex TUI Defaults Local sessions leave reasoning summaries off by default while still honoring explicit user settings.

What this means for you

For Vibe Builders: You can now run multi-week coding projects inside Cursor Projects without managing Git repos or servers. Claude Code and Codex CLI voice and sync features let you steer agents from the terminal or phone while keeping context across sessions. Start with Cursor scratch projects to test a full app build this week.

For Non-techies: For your business, Cursor now handles deployment checks and security scans automatically so you spend less time reviewing code. Claude Code and Codex updates improve reliability when you connect agents to Slack or cloud services. Test one new agent workflow on a small internal task before scaling.

For Developers: Cursor self-hosted machines and Projects give you control over secrets and long-running agents that you can compare against Codex daemon recovery and Claude gateway proxy settings. Evaluate Bedrock integrations across all three tools for production reliability. Benchmark isolated subagent performance against your current CI setup this week.

What to watch next

Watch for Cursor Origin rollout to more plans and any Codex GPT-6 model benchmarks. Check Claude Code gateway updates for Enterprise proxy controls. Track agent task recovery rates across the three tools.