# New Ways of Working with Codex — Dominik Kundel & Charlie Guo, OpenAI

Source: [AI Engineer](https://www.youtube.com/watch?v=IPh9YsWvDQk)  
Feed7 permalink: https://feed7.dev/p/new-ways-of-working-with-codex-dominik-kundel-charlie-guo-openai-04dt3lk  
Published: 2026-11-17T19:43:56.000Z  
Trust: Source Linked (source_linked)

## Why Included

The talk shows Codex extending across local, cloud, browser, and computer-use workflows, with independent chats and code review in the app. Builders can choose where tasks run and how agents gather context.

## Source Summary

The speakers describe Codex work across local machines and **Codex Cloud**, plus background computer use on macOS and a Chrome extension for browser context on Windows, Linux, and macOS. The talk also covers independent chats, CLI session switching, voice mode, and an in-app code review tab.

## Practical Implication

For long-running work, consider whether tasks belong on a local dev box or in a cloud environment. Use computer or browser access when an agent needs information from an app or logged-in site, and independent chats when work needs separate steering.

## Agent-Ready Context

The speakers describe Codex work across local machines and **Codex Cloud**, plus background computer use on macOS and a Chrome extension for browser context on Windows, Linux, and macOS. The talk also covers independent chats, CLI session switching, voice mode, and an in-app code review tab.

For long-running work, consider whether tasks belong on a local dev box or in a cloud environment. Use computer or browser access when an agent needs information from an app or logged-in site, and independent chats when work needs separate steering.

Computer history is macOS-only in the talk and must be enabled before it begins building context. The speakers give feature demonstrations, not comparative reliability or performance measurements.

## Connected Context

Feed7 judgment across 1116 accumulated Signals:

The talk describes Codex across local and cloud work, with computer and browser context, independent chats, session switching, voice, and code review. It clarifies that computer history must be enabled before context builds and that the demonstrated macOS history feature and workflows are not comparative evidence of reliability or performance.

- [Run Paseo agents on Vercel Sandbox](https://feed7.dev/p/run-paseo-agents-on-vercel-sandbox-1pvz431) — Paseo makes cross-device continuation concrete and specifies what persists when its sandbox stops, adding a lifecycle detail to the talk’s broader local-versus-cloud task choice.
- [citrolabs/ego-lite](https://feed7.dev/p/ego-lite-1i7kk2j) — ego-lite provides isolated logged-in browser spaces on macOS, a concrete approach to the authenticated browser context discussed in the talk, with access boundaries that need consideration.
- [openai/codex](https://feed7.dev/p/codex-1jzl221) — The Codex repository describes the shared agent across terminal, IDE, desktop, and web surfaces, complementing the talk’s examples while leaving differences between those surfaces unresolved.

## Context Map

- Layer: tools
- Domains: coding
- Topics: coding-agents, cloud-agents, computer-use

## Uncertainty

- Computer history is macOS-only in the talk and must be enabled before it begins building context. The speakers give feature demonstrations, not comparative reliability or performance measurements.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.
