Cursor agents get read-write access to Gmail, Drive, and Calendar

Your coding agent can now send email and edit your calendar — real reach, real blast radius. Plus Vercel traces every LLM call and GitHub Spark is retiring.

Nowline AUG 7 1:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Your agent, now inside your inbox and calendar

    Cursor's new Google Workspace plugins give agents direct read-write access to Gmail, Drive, and Calendar — search and send mail, manage threads and labels, create and organize Drive files, and read or edit calendar events, all without leaving the editor. Install from the Cursor Marketplace or the Customize page. The catch: this is the “lethal trifecta” by design — private data, the power to act, and untrusted content in one loop — so scope each service tightly and keep a human on the send button.

  • Vercel now traces every model call

    AI Gateway can now emit an OpenTelemetry trace for every request and ship it through Vercel Drains to Braintrust, Sentry, Dash0, Statsig, or any OTLP endpoint. Each trace carries model routing, retries and fallbacks, token counts and cost, time-to-first-token and request duration — prompt and completion text stay out for privacy. Pro and Enterprise teams get real per-call observability without bolting on a separate SDK.

  • New integrations now teach your agent to use them

    Run `vercel integration add neon` (or any Marketplace provider) and the CLI now also pulls that provider's agent skills from skills.sh, so coding agents like Claude Code, Cursor, and v0 know the correct API surface on the first try instead of guessing. It's a small quality-of-life win that trims the “agent hallucinated the SDK” tax every time you wire up a new service.

  • GitHub Spark is shutting down — export by Aug 31

    GitHub stopped new Spark signups and app creation on Aug 4, and the service fully shuts down Aug 31. Apps you've already deployed keep running, but export your code (··· → Create repository) before then if you want to keep editing, and anything calling `llm()` needs your own inference provider now that GitHub Models is already retired.