Cursor and Claude Code Roll Out New Agent Controls
TL;DR
Cursor added inline reviews and token tracking while Claude Code improved stability and plugin loading; OpenAI Codex brought Vim editing and a Chrome extension, giving builders clearer ways to manage AI coding agents across daily workflows.
What shipped
On 08 May the main AI coding tools released targeted fixes and features for agent coordination. Cursor focused on review speed and spend visibility. Claude Code addressed connection issues and remote rules. OpenAI Codex added keyboard-first editing and browser support.
Cursor
Cursor expanded its agent features to handle reviews and team settings in one place. The updates target both individual builders tracking usage and larger groups sharing custom setups. These changes reduce the need for extra tools when coordinating edits or controlling costs.
- •Inline PR Reviews Cursor introduced inline pull request views with async subagents that handle parallel edits. Teams now complete reviews quicker and sync changes across branches more smoothly than with standard Git flows.
- •Token Dashboard Cursor released a breakdown of token use per agent. Builders spot bloated rules or MCP setups early and cut waste before monthly costs rise.
- •Admin Controls Cursor added model blocklists, soft spend caps, and surface analytics for enterprise admins. Security teams monitor agent activity across reviews and tasks without extra monitoring software.
- •Shared Plugins Cursor enabled teams to publish custom rules and MCP servers directly in settings. Groups share skills internally without separate Git repositories for each project.
Claude Code
Claude Code focused on connection reliability and remote options this cycle. Fixes cover worktree handling, Windows paths, and plugin loading from URLs. The updates help agents stay active during multi-branch work and adjust effort based on task load.
- •Worktree Fixes Claude Code improved worktree handling and fixed mTLS plus proxy errors. Agents maintain connections across branches without repeated login failures on multi-branch tasks.
- •Session Stability Claude Code added session ID variables and terminal flags to prevent screen takeovers. SIGINT handling now allows clean exits instead of hangs during interrupts.
- •Windows Activation Claude Code corrected path errors for the VS Code extension on Windows and fixed API key headers to the Mantle endpoint. Developers activate tools on standard Windows setups without custom fixes.
- •Remote Plugins Claude Code introduced remote plugin loading and effort-tracking hooks. Agents pull rules from URLs and adjust behavior according to task load.
OpenAI Codex
OpenAI Codex added keyboard controls and a lightweight browser helper. The CLI now supports modal Vim editing with workspace rules. The new extension lets agents assist across tabs without taking over the full window.
- •Vim Editing in TUI Codex CLI added modal Vim editing and workspace-aware plugin controls in the text interface. Power users edit code precisely while keeping access rules per workspace.
- •Chrome Extension Codex launched a Chrome extension that operates across multiple tabs. Builders get web task help without the agent claiming the entire browser window.
What this means for you
For Vibe Builders: You can now manage agent reviews and token spend inside Cursor without switching tools. Claude Code stability fixes mean fewer login interruptions when running parallel tasks. The Codex Chrome extension lets you delegate web work while keeping control of your tabs.
For Non-techies: For your business these updates mean AI coding tools now track their own usage and share rules across a small team. Cursor spend caps and admin views reduce surprise bills. Claude Code remote plugins let you load fixes without local setup hassles.
For Developers: On the platform side Cursor token analytics and shared MCP servers give clearer signals for production rules. Claude Code session and SIGINT fixes improve reliability in multi-branch CI flows. Watch the new Codex TUI Vim mode for local editing benchmarks against your current setup.
What to watch next
Track whether Cursor publishes its token dashboard metrics publicly. Watch for Claude Code updates on remote plugin adoption rates. Note any Codex extension feedback from multi-tab workflows in the coming days.
Harsh’s take
The pattern across these releases shows vendors adding visibility and control layers rather than raw model gains. Cursor and Claude Code both prioritize spend tracking and connection fixes over new capabilities. This suggests the real bottleneck has shifted from model power to operational overhead in agent use.
A contrarian view is that these controls may slow down solo builders who just want quick edits. Yet the second-order effect is clearer cost signals that help teams avoid waste. Builders should test the new Cursor token dashboard against their current agent rules this week and remove any obvious bloat before the next billing cycle.
by Harsh Desai
More AI news
- Daily RoundupVercel Flags and WebSockets, Google Interactions API, and agent tools for live apps
Vendors released feature flags, WebSocket support, unified model APIs, new video models, trending OCR tools, and agent deployment options on 22 June, giving builders direct paths to ship realtime and segmented AI features.
- FeatureLovable Build with URL links now reference public web pages
Lovable's Build with URL links can now reference public web pages alongside images. The feature uses the referenced page's layout, content, and styling to recreate or iterate on it.
- FeatureSet up cloud environments and run subagents with /in-cloud
Cursor's /in-cloud sets up cloud development environments in under 10 minutes and runs isolated subagents. Sessions hand off between local machines and the cloud.