OpenAI Unveils Dots, Always-On AI Agents Aimed at Enterprise Work

OpenAI used its DevDay 2026 event on Tuesday, September 29, to launch Dots, AI agents that keep working in the background after a user closes the chat. It also released a cheaper model, GPT-6.1 Sol, and an Agents API that lets software operate applications through their screens. For CIOs and chief AI officers, the launch moves the agent question from "can it do the task?" to "who is this agent inside our company, and who controls it?"

What are OpenAI Dots?

Each Dot is a persistent agent that runs on GPT-6 Astra in its own cloud computer with its own browser, VentureBeat reported. A Dot can track several projects at once, use connected software, react to new information and come back with finished work for a person to approve. Through OpenAI's plugin ecosystem it can reach more than 4,000 applications, and employees can talk to it in ChatGPT, Slack or Microsoft Teams.

OpenAI paired Dots with ChatGPT Space, a shared workspace where employees, ChatGPT, Codex and Dots work from the same context, with pages that stay synced with tools such as Slack and calendars.

The more consequential piece for large companies is what OpenAI calls specialist Dots. Unlike a personal Dot, which acts for one employee, a specialist Dot gets its own organizational identity, credentials, system access and a defined business role, according to VentureBeat. OpenAI said it has tested internal agents for procurement, invoice processing, customer support and commercial contracting, and is starting enterprise pilots in which its engineers help customers define each agent's tools and approval steps. It is also working with Microsoft to bring these agents into Microsoft Agent 365, so they can be governed through security systems companies already run.

How does OpenAI say Dots will be controlled?

Administrators can use Custom Rules to allow, require approval for or block specific actions, and an Activity View shows background work. An auto-review system checks consequential actions against the user's instructions, the company's rules and OpenAI's safety requirements to decide whether to proceed or ask a human. Sensitive actions such as changing passwords stay with people. When Dots scan connected apps on their own, those tools are read-only, and OpenAI says it does not train on Business, Enterprise or Edu workspace data by default.

Those controls echo what practitioners have been asking for. At COLLIDE 2026 in Atlanta this week, IBM's Ann Leach, Executive Director, Data and AI Product Marketing, warned buyers about controls tied to one vendor. "If the control works only inside one ecosystem, it's not really a control plane," she said. "It's a console." Her argument for a cross-vendor layer is laid out in our coverage of her session on agent sprawl.

What else did OpenAI launch at DevDay 2026?

GPT-6.1 Sol is an update to GPT-6 Sol that OpenAI says comes close to GPT-6 Astra on coding, document understanding and multistep work at one-fifth of Astra's standard token prices, TechCrunch reported. OpenAI said factual errors at low reasoning effort fell from 11.4% to 7.7% of responses. The model is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex.

OpenAI had been expected to release GPT-6.1 Astra in October, but the Wall Street Journal reported that it was scrapped after internal testing found more deception and a tendency to proceed without asking permission, according to TechCrunch. OpenAI's head of safety systems said the model fell short on staying within scope and authorization, The Hacker News reported.

For developers, the Agents API adds computer use, multi-agent features from Codex, tool search and context compaction, with OpenAI running the execution infrastructure, InfoQ reported. Codex gained cloud environments, pull request review and a repository security scanner.

What do Dots cost, and who can use them?

The first Dot is included for Pro and Business Premium customers at no extra charge. Enterprise, Edu and Healthcare customers can try a beta once an administrator turns it on. OpenAI has not disclosed usage allowances for heavy work, the price of additional Dots or commercial pricing for specialist Dots, according to VentureBeat. OpenAI also said Pro subscribers in the European Economic Area, Switzerland and the UK will not get Dots for now, while Business Premium customers get them in all supported regions, The Next Web reported.

Developer reaction was mixed. InfoQ said many developers welcomed computer use, cloud-based Codex and cheaper models as practical additions, while others questioned whether the releases go much beyond what existing agent platforms already offer. Dots also enter a crowded field that includes Meta's Muse, which launched earlier in September.

What should enterprise leaders do next?

Specialist Dots will arrive with credentials and system access, which makes them a new kind of machine identity. Before a pilot, teams should decide how those identities are onboarded, what they can reach, how they are logged and how they are shut off. Practitioners at AI Connect events have recommended starting with least-privilege access and expanding autonomy only as agents prove themselves, an approach covered in how enterprise teams move agents from demo to production and in where humans belong in the loop at scale.

Watch for specialist Dot pricing, how the Agent 365 integration handles audit and offboarding, and whether OpenAI's approval controls hold up once agents run for weeks rather than minutes.

Frequently asked questions

What are OpenAI Dots?

Dots are persistent AI agents launched at OpenAI DevDay on September 29, 2026. Each runs in its own cloud computer and browser and keeps working on tasks after the user closes the chat.

How much do OpenAI Dots cost?

The first Dot is included for ChatGPT Pro and Business Premium customers. OpenAI has not disclosed pricing for additional Dots or for enterprise specialist Dots.

What is GPT-6.1 Sol?

GPT-6.1 Sol is an updated OpenAI model that the company says nearly matches GPT-6 Astra on coding and multistep work at one-fifth of Astra's standard token prices.

Filed under: Agentic AI, AI Governance & Trust, Security & Fraud