En 30 segundos
An OpenAI presenter demonstrates Codex cloud environments, private resource access, collaboration in Slack and Spaces, personal dots, and updated CLI and web interfaces. The talk includes a failed live step and distinguishes planned features from the demo.
- 1Longer tasksCloud execution reduces reliance on a continuously running personal machine.
- 2Cloud setupCodex proposes setup artifacts after inspecting selected repositories.
- 3Design to demoA task uses a Figma design to implement and test a shader.
- 4Private resourcesThe presenter demonstrates private network access with Tailscale.
- 5Team reviewChatGPT and shared environments support collaborative pull request work.
- 6Shared plansA Space keeps plans and progress together.
- 7Personal dotA dot can track context and delegate work within its permissions.
Conclusiones
- Cloud execution reduces reliance on a continuously running personal machine.
- Codex proposes setup artifacts after inspecting selected repositories.
- A task uses a Figma design to implement and test a shader.
- The presenter demonstrates private network access with Tailscale.
- ChatGPT and shared environments support collaborative pull request work.
- A Space keeps plans and progress together.
- A dot can track context and delegate work within its permissions.
Lee el texto editado por capítulos
Why cloud execution
Longer-running delegated work.
At OpenAI DevDay 2026, a presenter who introduces himself as Dominic demonstrates how Codex can work across cloud environments, collaboration tools, the CLI and the web. The presentation describes features and plans as of October 2026; availability can change.
The presenter describes a shift from delegating one or two tasks to running several agents and handing off work that can take hours or days. Keeping a laptop or a separate machine running adds maintenance, which motivates a cloud option.
Cloud environment demo
Setup, design and private access.
In the desktop demo, a cloud environment is created with selected repositories. Codex inspects dependencies and proposes a startup script and skill; the Linux environment can edit iOS source but cannot run the iOS app. The presenter says bring-your-own cloud environments are planned for later.
A separate task uses a Figma plugin to implement a shader from a design, then produces an interactive site for testing adjustments. The presenter shows how to inspect changes and leave comments before using the result.
The demonstration connects a cloud environment to a private Raspberry Pi API through Tailscale and mentions OIDC for cloud resources. The live task encounters a problem, so the presenter shows a previously run example instead.
Collaboration and review
Slack, pull requests and shared plans.
In Slack, ChatGPT is used to inspect a pull request and a design, then test changes in a shared cloud environment. The desktop code review view brings assigned and authored pull requests together and supports comments, review and questions.
The presenter shows a Space holding a shared plan, diagram and task tracker. Codex can continue to update the document as the work progresses.
Dots, CLI and web
Ways to track and delegate work.
The presenter describes a personal dot that can monitor discussions and delegate a deeper review to a cloud task. In this demonstration, the dot is associated with a person’s context and permissions; @ChatGPT is presented as a shared collaboration option using shared cloud environments.
The CLI demo includes an agent view for tracking multiple tasks, plugin mentions, terminal rendering and voice. A browser view provides access to cloud tasks and the dot; the presenter describes web code review as coming later.
The closing suggestions are to try cloud environments for longer tasks, use a shared Space for plans and progress, and give a dot an ongoing responsibility when appropriate.
Más de este canal
34:50EN
OpenAI· 7 oct 2026ClawLabs: Building Open Source with Shared AgentsKevin LinA DevDay demonstration of team workflows, ongoing automation and agent controls.
18:35ENOpenAI· 7 oct 2026Itamar Friedman on AI Code Review and Engineering ContextDanielle, Itamar FriedmanQodo’s CEO discusses reviewers, knowledge and human judgment at DevDay 2026.
23:00ENOpenAI· 7 oct 2026From Individual Agents to Team Workflows: A DevDay DemonstrationJason · Jane StreetOpenAI’s Jason explores context, shared pages and reusable plugins.
23:04ENOpenAI· 7 oct 2026AI Cost per Completed Task: Models, Caching and EvaluationAn OpenAI DevDay panel discusses cost–quality tradeoffs in AI applications.
19:40ENOpenAI· 7 oct 2026How OpenAI Teams Use ChatGPT for Ongoing WorkA DevDay 2026 presentation on launch monitoring, research and shared knowledge.
12:49ENOpenAI· 7 oct 2026Alex Mashrabov: Higgsfield agents that choose models and launch their own GPUsAlex Mashrabov · HiggsfieldAt OpenAI DevDay 2026, Higgsfield's CEO walks through why a continuous media experience needs an agent that picks models, scores its own output, and provisions GPUs — including a live 10-hour AI stream small teams can now ship.