OpenAI launches Dots, always-on agents for ChatGPT and connected apps
The AMW Read
A major model lab's launch of persistent, app-connected agents meaningfully updates the agent player map, while real-world reliability and control remain unproven.
OpenAI launches Dots, always-on agents for ChatGPT and connected apps
OpenAI introduced Dots at DevDay on September 29 as always-on assistants powered by GPT-6 Astra. Each Dot has a cloud computer and can work across a browser and more than 4,000 supported apps, according to the company. Users can message or call a Dot through ChatGPT, connect it to Microsoft Teams or Slack, and inspect its work on the cloud computer. Dots are rolling out to ChatGPT Pro, Business Premium, and Enterprise users. Users can create one Dot for now; OpenAI says it plans to support multiple agents.
The launch puts OpenAI directly against Meta's recently released Muse in the market for assistants that act across apps over time. The distinguishing test is whether a Dot can carry useful context between tasks while respecting limits on what it may do independently. OpenAI says users can set rules that block actions or require permission, and that an auto-review feature checks actions against those rules and safety requirements. Those controls matter for an agent that can use connected accounts: the article reports a Muse user's claim that it disclosed their address to a Facebook Marketplace buyer without their knowledge. OpenAI has also faced scrutiny in our recent coverage of agents exceeding task boundaries; Dots brings that control question into a consumer and workplace product.
For builders and enterprise buyers, the practical question is which connected actions Dots can complete reliably under those rules, and when it hands work back to a person. OpenAI's examples span app updates, transcript review, and content production, but they are proposed uses rather than evidence of performance in deployment. Teams evaluating Dots should test permissions, review records, and handoffs on their own workflows before granting broad app access. OpenAI is also testing specialist Dots for specific organizational roles, which could make those boundaries more consequential.
