Meet the New Codex CLI

Summary
Craig Dennis, Developer Experience at OpenAI, presents an overview of major new updates to the Codex CLI. He demonstrates the new full-screen terminal interface, native voice interaction mode, multi-agent background task execution, worktree branching, mobile remote control, terminal usage analytics, and terminal rendering for Mermaid diagrams and LaTeX.

What is shown

• [00:25] A Python CLI project (burrito-me) for automated DoorDash ordering, followed by its companion web front-end burrito.me [00:33].
• [00:40] The new full-screen Codex CLI interface running GPT-6-Astra default fast, featuring scrollable message history and a fixed input footer.
• [00:54] Initiating two-way voice mode (/voice), speaking directly to Codex to spawn three distinct agent tasks simultaneously: adding a repeating burrito background to the web app, testing the CLI for price changes/order confirmations, and generating a Mermaid diagram of the order flow.
• [01:26] The Agent Command Center view, demonstrating multi-agent monitoring, filtering by status, grouping, searching, and inspecting individual task output.
• [01:54] Forking conversation threads into isolated git worktrees (/fork, /new) to pursue alternate implementations in separate checkouts while preserving context.
• [02:07] Remote control via the Codex mobile interface, showing live task inspection, diagram reviews, and mobile prompt input synced to the laptop session.
• [02:26] The updated web front-end displaying the newly generated repeating burrito pattern background.
• [02:32] Built-in usage analytics dashboards in the terminal displaying lifetime token stats, peak tokens, active streaks, top chats, plugins, and skills.
• [02:41] Inline terminal rendering of Mermaid flowcharts and formatted LaTeX equations.
• [02:56] A terminal summary graphic outlining commands for worktrees (/worktree, /fork, /new) and full-screen UI controls (F3, /theme, /tui, /copy, Esc twice).

Claims & numbers

• The presenter says users can fork conversations into a new worktree to retain identical context in a completely separate checkout [01:57].
• The usage analytics dashboard displays metric examples including 79.2M lifetime tokens, 35M peak tokens, a longest chat of 1h 59m, and 4-day streaks [02:33].
• The presenter claims users can enable remote control from a laptop and interact with running Codex tasks directly from their phone [02:10].

Notable quotes

• "Look, we know some of you are terminally in the terminal. Well, this one's for you all." [00:03]
• "You want Codex to feel fast and stay out of your way. You want it to let you build where you love to hang out." [00:10]
• "And if I want to try to build in a different direction, I can fork this conversation into a new worktree." [01:53]

Assessment
This is an official OpenAI launch and developer demonstration video featuring genuine CLI screencasts mixed with presenter presentation. While the workflow demonstrates real tooling features across terminal and mobile clients, the underlying code execution and agent completions are fast-forwarded and edited for presentation timing.

Described by gemini-3.8-flash on 2026-10-05 from the video's audio and frames.