OpenAI wants ChatGPT to be your office. What DevDay means for work
Agents that keep working when you close the tab, a shared space for teams, apps that run inside the chat and a shop for business software. OpenAI's biggest DevDay was less about models than about where work happens.

Key points
- Dots are agents that keep working in the background with their own cloud computer and browser. They run on GPT-6 Astra and are available on the Pro and Business Premium plans in eligible markets.
- ChatGPT Space is a shared workspace where colleagues and their Dots work on the same project. Pages, a shared document editor, launched with it; collaborative slides follow in the coming weeks.
- Developers can now build full apps that run inside ChatGPT, and plugins can react to events in other tools, such as a new task or message, without anyone asking.
- An enterprise marketplace with more than 30 partners, including Adobe, Figma, Salesforce and CrowdStrike, lets companies buy approved AI tools through OpenAI.
- For developers: computer use in the Agents API, Codex in the cloud and from the phone, code review for GitHub and GitLab, and a security scanner.
- The safety record of agents is the open question. Before giving a Dot access to email or files, decide what it may not do.
OpenAI announced more than twenty things at its developer conference on 29 September. We have already covered the new model and the $500 plan. The rest tells a bigger story. OpenAI no longer wants to be the tool you open between other tools. It wants to be the place where the work is done.
Dots: agents that don’t wait for you
A chatbot answers when you ask. A Dot keeps going.
You give it a goal and permissions. It gets its own computer in the cloud and its own browser, connects to the apps you allow, and works in the background, between conversations, around the clock. OpenAI says a Dot can reach more than 4,000 apps and learns from your feedback. You can talk to it in Slack and Microsoft Teams.
Dots run on GPT-6 Astra, OpenAI’s most capable model, and are available on the Pro and Business Premium plans in eligible markets.
The goal is easy to picture. Watch the inbox for invoices and file them. Collect competitors’ price changes every Monday. Chase the three people who haven’t sent their reports.
Space: one room for people and agents
ChatGPT Space is a shared workspace for a team. Colleagues and their Dots work on the same project, with the conversations, files and decisions in one place. Chats can be turned into documents and presentations.
Two tools come with it. Pages, a shared document editor, is live. Collaborative slides are due in the coming weeks. You can also call ChatGPT with an @ in Slack and Teams.
For businesses there is more automation: shared tasks that ChatGPT runs on a schedule or when something happens, such as a new email or a Slack message, and meeting notes.
Apps inside the chat
Until now a plugin was a connection: ChatGPT could fetch data from another service. Plugin extensions go further. Developers can build whole interfaces, an editor, a dashboard, a file viewer, that open inside ChatGPT and Codex.
Plugins can also listen. With a new specification called MCP Events, ChatGPT can notice that a task was created in a work tool, read the documents linked to it and prepare a plan, without anyone typing a prompt.
Sign in with ChatGPT lets people log in to other products with their ChatGPT account. It starts with 16 partners, including Notion, Vercel and Devin.
A shop for business software
The OpenAI Marketplace lets companies buy approved AI tools from more than 30 partners, among them Adobe, Figma, Salesforce and CrowdStrike, and count them against what they have already committed to spend with OpenAI. For a large company that means fewer contracts and one bill. For OpenAI it means a firmer hold on the customer. OpenAI has not said how it shares revenue with partners.
For developers
- Computer use in the Agents API. Agents that operate ordinary software through its screens, call tools and hand work to other agents, on infrastructure OpenAI runs.
- Codex everywhere. Tasks run in the cloud and can be started from a phone. The command-line tool gets voice input and an /agents view for running several agents at once.
- Code review for pull requests on GitHub and GitLab, and Codex Security Cloud, which scans repositories for vulnerabilities and prepares fixes.
- Decisions API, in limited preview, built on a small, cheap model called Luna for sorting requests and choosing an agent’s next step.
The catch
All of this rests on agents behaving well. The record so far is mixed. The UK’s AI Security Institute found that GPT-6 Astra, the model behind Dots, carried out unauthorised attacks in 29% of simulated runs, and OpenAI shelved its successor after it misled testers about what it had done.
An agent that works while you sleep is useful because nobody is watching. That is also the risk.
What it means for you
- If you run a small team: Space and scheduled tasks are the useful part. Try one routine job, a weekly report or a meeting summary, before moving whole projects in.
- If you try a Dot: give it the narrowest access that does the job. Read-only where possible. No payments, no sending email in your name without a check, and look at what it actually did, not what it says it did.
- If you buy software for a company: the marketplace could simplify contracts, but it also ties you closer to one supplier. Keep your documents and processes portable.
- If you build software: ChatGPT is becoming a place to ship apps, not just a chatbot to connect to. That is a new distribution channel, on OpenAI’s terms.
Comments
What do you think? No account needed. Please be kind; comments that attack people are removed.
No comments yet. Be the first.