Talk to Cursor, keep Codex tasks in view

Originally published in The Vibe Check.

🛠️ Get the Vibe Check Agent Skills Pack ($12): 12 ready-to-use agent skills + PDF

Catch-up edition for Nov 7, 2026; prepared Nov 8, 2026 (NZDT). Sent late, not backdated.

What changed in AI agents and coding tools, and whether it matters to you. 90 seconds, no hype.

---

1. Cursor voice is now a conversation, not just dictation

The November 6 update adds real-time calls with agents in the Agents Window, IDE, web and iOS. You can ask for progress while the coding agent works and hear short updates. Cursor says voice chat runs on Grok Voice, with zero data retention when Privacy Mode is on.

Does it matter to you? Useful when you want to steer work without watching a terminal. The waveform starts a call; dictation still just turns speech into a written prompt. Check Privacy Mode before discussing sensitive work.

Source: Cursor Changelog

2. Codex 0.167 adds a terminal agent command center

Run codex agents --no-daemon to open the command center. It and the session picker mark unread results. You can switch connected backends without losing drafts, and supported terminals can preview local image attachments while you compose.

Does it matter to you? Running several tasks is easier when finished work doesn’t disappear into tabs. The release also keeps saved chats visible across model-provider switches; if one can’t resume, you can still read its transcript.

Source: Codex release notes

3. Claude Code can deny instead of waiting forever

v2.1.319 adds permissions.denyInsteadOfPrompting: actions that would wait for approval are refused and the turn continues. MCP tasks can also move into the background and report back when finished. Auto mode changes when telemetry or feature flags are off: uncheckable calls are asked about or denied rather than silently using the built-in classifier.

Does it matter to you? Unattended runs should fail visibly, not hang or bypass checks. The release also fixes several permission bypasses and a worktree sandbox escape. Update before relying on the new behavior.

Source: Claude Code release notes

---

Try this today

On a disposable task in Claude Code, test what happens when an action needs permission and no one is there to answer. Verify the run’s final status and output, not just whether the process stopped. Keep your existing deny rules.

---

Hit reply and tell me which tool you want tested next.

Want your agent to do more out of the box? Grab the Vibe Check Agent Skills Pack.

Know someone who’d find this useful? Forward it, or send them to The Vibe Check.