ChatGPT dots can start Codex tasks using conversation context

• Dots can start Codex tasks and follow up on existing threads using conversation and automation context, according to ChatGPT’s announcement.
• Dot creation now works directly in the ChatGPT app on iOS and Android, as ChatGPT announced.
• Dots can review and edit Scheduled Tasks in ChatGPT Work, including recurring work, per the delegation announcement.

Dots, OpenAI’s always-on ChatGPT agents, have their own cloud computer, and threads they manage can rename themselves as work changes. A quieter fix stops oversized Gmail previews from hiding approval buttons.

OpenAI’s October 9 notes describe four capabilities:

1. Start work in Codex.
2. Follow up on existing Codex threads.
3. Search ChatGPT conversations and use context from existing Codex threads and automations.
4. Decide whether to continue an existing thread or start a new one.

Thread continuity is the standout upgrade for long-running coding projects.

Dots can inspect scheduled work and change recurring tasks in ChatGPT Work, as ChatGPT’s announcement describes.

Before the update, a user reported on October 7 that a Work conversation could retrieve existing scheduled tasks their dot could not find. They wanted to preserve those tasks rather than recreate them and risk duplicate runs or alerts.

Dots started on desktop for power users; creation and customization now work inside ChatGPT’s iOS and Android apps.

The mobile setup notes cover:

• Giving the dot a name and customizing its appearance.
• Connecting plugins from the phone.
• Choosing to open ChatGPT directly into the dot’s conversation.

OpenAI’s launch rollout specifies these access boundaries:

• Pro and Business Premium: gradual availability for eligible users aged 18 and older in supported markets.
• Pro: the European Economic Area, Switzerland and the UK are excluded at launch.
• Enterprise: beta access, off by default.

• Personal app fixes: dkundel sends messages to his dot to request changes in his macOS utility app. He lets it “push straight through” because he is the only affected user, as he explained in a reply.
• Email bids: his dot began comparing construction bids automatically as they arrived in his email.
• Meeting corrections: an OpenAI colleague’s dot DM’d them a correction they could share during a meeting, according to embirico’s account.

Separately, tldraw’s ChatGPT plugin lets users @mention ChatGPT or Codex from a canvas to prompt the agent. The agent can read and respond to comments directly on the canvas.

Instructions in agents.md can ask the agent to use sticky notes, according to max__drake’s reply.

OpenAI also shipped these reliability fixes:

• Browsing and scrolling: faster cloud browsing and sidebar loading, fewer pauses awaiting replies, smoother scrolling and a fix for excessive network usage.
• Notifications: fewer extra phone notifications while ChatGPT is already in use.
• Conversation rendering: better code blocks and attachments, plus character-rendering fixes in Safari and Firefox.
• Setup: a fix for Outlook setup getting stuck and autofill for email verification codes.
• Recovery: clearer plan guidance and retries for activity that fails to load.
• Enterprise: fixes for web access and starting voice calls.