OpenAI Launches Dots, Always-On Agents to Challenge Meta’s Muse
OpenAI’s Dots bring always-on AI agents to ChatGPT, Slack, and Teams, escalating the race with Meta’s Muse over autonomous workplace automation and control.
OpenAI has introduced Dots, always-on AI agents designed to monitor ongoing work, perform tasks on their own cloud computers, and contact users when they need attention. The product, announced at the company’s DevDay 2026 developer conference, puts OpenAI directly into competition with Meta’s recently launched Muse agent.
Dots are intended to move beyond the current model of starting an AI task one request at a time. According to reporting from The Decoder, users can assign a Dot research, data analysis, document creation, coding, and other work, then allow it to continue operating in the background. The agents can be reached through ChatGPT, Slack, and Microsoft Teams, with OpenAI saying they retain context across those channels.
The launch matters because autonomous workplace software is shifting from a chatbot feature to a persistent operational layer. For builders and enterprise buyers, the key questions are no longer only whether an AI model can complete a task, but whether an agent can manage permissions, preserve context, work reliably without constant supervision, and stop before taking a consequential action.
Dots are designed to keep working after the prompt
Each Dot runs on its own cloud computer with access to a browser, terminal, and file storage, according to The Decoder’s account of OpenAI’s announcement. The agent can conduct research, analyze data, draft materials, or build software through Codex. OpenAI’s demonstrations reportedly showed a desktop interface with the conversation on one side and a view of the agent’s active work on the other.
That architecture distinguishes Dots from Computer Use in ChatGPT Work, where a user starts each job and the agent operates on the user’s screen. A Dot instead works in a separate environment by default. Users can step in through a “Take over” control, while access to a personal laptop requires permission.
OpenAI says Dots can also connect to more than 4,000 applications through plugins. The company’s longer-term plan is for teams of Dots to work together, although the initial launch provides one Dot per user. Business customers receive broader options, including specialized agents with their own identities, credentials, and, where required, company-managed computers.
The Decoder reported that OpenAI has tested specialized Dots in purchasing, invoice processing, email marketing, customer support, and contract management. The rollout begins with pilots at individual companies, and OpenAI is working with Microsoft to bring those specialized agents to Agent 365, Microsoft’s management platform.
Safety controls are central to the product pitch
OpenAI’s proposed operating model separates background discovery from actions that can change systems or share information. When a Dot is not actively working with a user, its “proactive research” mode is limited to read-only tools, according to The Decoder. In that mode, it cannot send messages, edit content, or control a browser or computer.
For active tasks, users can create Custom Rules that permit specific actions, require approval, or prohibit them. OpenAI says an automated review checks account-related or information-sharing actions against those rules. Certain sensitive operations, including password changes, remain human-controlled. The company also says a Dot can use stored passwords without exposing them to the model.
These safeguards are vendor-reported design claims rather than independently verified evidence of reliability. OpenAI acknowledges that Dots can make mistakes and recommends checking results that could have real-world consequences. The company also says monitoring systems can pause or stop an agent when safety concerns arise, while an Activity View shows active and background tasks.
The product’s access and data policies are similarly segmented. OpenAI says it does not train on content from Business, Enterprise, and Edu workspaces by default. Personal-plan users can choose whether their Dot conversations and work are used for training. The company also says background research and the agent’s internal notes are not used directly for training.
A direct response to Meta’s Muse
The timing makes the competitive comparison difficult to miss. The New York Times, WIRED, Yahoo Finance, and qz.com all framed the launch as OpenAI’s answer to Meta’s Muse, while The Decoder reported that the two products share a similar core concept: an agent operating on a secured computer with a browser, working in the background, and requesting approval for sensitive actions.
The available coverage does not provide independent benchmarks comparing Dots and Muse on task completion, latency, cost, or safety. Nor does the source material establish adoption figures for either product. Claims about internal tests, early users, or demonstrated workflows should therefore be treated as company-provided examples rather than proof that the agents can reliably manage general business operations.
OpenAI is also positioning Dots as part of a broader product stack. The agents reportedly run on GPT-6 Astra, while OpenAI is releasing the cheaper GPT-6.1 Sol model. The Decoder reported that the top-end GPT-6.1 Astra was being held back because of safety concerns. Those model details come from specialist reporting on the event and were not accompanied in the supplied evidence by independent technical evaluations.
What Dots mean for builders and enterprises
For product teams, the significant change is persistence. A conventional assistant responds to a prompt; a Dot can watch a project, retain context, and decide when a follow-up is useful. That could reduce coordination work in areas such as bug triage, invoice preparation, recurring research, and document production. It also creates a larger failure surface: stale context, incorrect assumptions, duplicate actions, and unauthorized data movement can accumulate while nobody is watching.
The approval model will be especially important for enterprise deployment. Read-only background work may be easier to authorize, but the value of workplace automation usually comes from changing records, sending communications, updating systems, or making purchases. Buyers will need clear audit trails, granular credentials, environment isolation, rollback options, and controls that administrators can enforce independently of an individual user’s instructions.
The cross-channel design may also determine whether Dots become useful or confusing. Reaching the same agent through ChatGPT, Slack, and Microsoft Teams could make adoption easier, but it raises questions about identity, conversation boundaries, notification overload, and which channel is authoritative when instructions conflict.
OpenAI’s planned specialized Dots point toward a marketplace for role-specific agents rather than a single general assistant. That could give companies more predictable workflows, but it may also increase governance costs as each agent receives its own permissions, credentials, and data access.
What to watch next
The first signal will be availability by customer type. The Decoder reported that one Dot is included with a subscription, while additional agents, speed, and work volume may cost extra later. Pro users in Europe currently face restrictions, while Business Premium customers have access in supported regions and Enterprise, Edu, and Healthcare users can join a beta when administrators enable it.
The next meaningful evidence will be independent testing of long-running tasks, error recovery, permission enforcement, and total operating cost. Enterprise buyers should also watch for details on retention, audit logs, administrator controls, plugin security, and the integration between specialized Dots and Agent 365.
Finally, competition with Meta’s Muse will be measured less by launch demonstrations than by sustained use. The decisive products will be the ones that can complete multi-step work accurately, explain what they did, and reliably ask for help before a mistake becomes expensive.
Creati.ai perspective
Dots show OpenAI betting that the next interface for AI work is not a chat window but an agent with time, memory, and a controlled computer. That is a meaningful product shift, but persistence increases the importance of governance at the same time it increases potential productivity.
OpenAI’s strongest opportunity is the combination of ChatGPT, Codex, Slack, Microsoft Teams, and enterprise administration. Its largest unresolved challenge is proving that autonomous work remains predictable outside a carefully staged demo. Until independent reliability and safety evidence appears, Dots should be evaluated as a promising automation layer—not as a replacement for human review.
