TLDR;
OpenAI has launched Dots, always-on agents that use their own cloud computers and keep working between conversations. They are rolling out to eligible Pro and Business Premium users, with an admin-enabled beta for Enterprise, Edu and Healthcare. Brands now need to govern persistent agents inside the company and expect more automated visitors outside it, including agents that can research products and make approved purchases with cards saved on merchant websites.
What happened
OpenAI announced Dots at DevDay on 29 September. Each Dot is powered by GPT-6 Astra, has a cloud computer and browser, and can use connected apps to pursue a goal over time. Users can communicate with it through ChatGPT, Slack or Teams, inspect its work and redirect it while a task is running.
The initial rollout covers Pro and Business Premium plans in eligible markets. Enterprise users, including Edu and Healthcare workspaces, can try the beta when an administrator enables it. OpenAI is also previewing specialist Dots with separate identities, credentials and access to company systems, starting with focused enterprise pilots rather than a general release.
TechCrunch described Dots as a packaged move toward independent agentic action rather than another chat interface. OpenAI’s examples include monitoring customer feedback, updating launch work as requirements change and revising analysis when new evidence arrives. Reuters framed the release as part of the competition to sell autonomous AI into enterprises, but availability and capability remain controlled by plan, market and workspace policy.
The product includes practical limits. Proactive background research uses read-only tools and cannot directly send messages, change connected content or control a browser. Other actions follow permissions, Custom Rules and a separate Auto-review system. OpenAI says Dots can make purchases with cards already saved on a merchant website, but each purchase requires user approval.
Why it matters
Dots change the expected lifespan of an AI task. A chatbot responds to a prompt and stops. A persistent agent can watch for changes, revisit a goal and act when a condition is met. For enterprise teams, that turns access design, approval policy and audit history into everyday operating requirements rather than a one-off integration exercise.
The external effect matters too. A Dot with a browser can arrive at a brand’s site to research, compare or buy. It may rely on product facts, policies, availability and checkout state without following the page in the order a human designer intended. Clear text, stable structured data and consistent transaction rules become service reliability for automated customers, not just search optimisation.
Risk grows with continuity. A misleading page, prompt injection or outdated permission can affect work that runs across several systems. Axios highlighted the safety pressure around launching always-on agents. OpenAI says Dots add isolated cloud environments, action checks and monitoring, but its own documentation still says they can make mistakes. Enterprise governance must assume useful autonomy and residual failure at the same time.
How your brand can benefit / be affected
Inside the business, start with narrow responsibilities and named owners. Give a Dot only the apps and data needed for one job, separate read access from write access and require approval for external messages, publishing, purchases and irreversible changes. Use a dedicated identity where possible so logs show what the agent did rather than attributing every action to an employee account.
OpenAI’s safeguards documentation makes permission design the practical control point. Test what happens when instructions conflict, a connected document contains malicious text, an approval arrives late or a task resumes after its context has changed. Review Activity View and downstream system logs together, and define how staff stop, correct and escalate a task that drifts.
For customer-facing journeys, test the site with an agent’s needs in mind. Keep prices, variants, delivery promises, returns and eligibility rules explicit. Do not hide material conditions behind hover states or visual-only controls. Make authentication and checkout handoffs predictable, and preserve source identifiers when an automated research journey becomes a sale or support case.
Measure outcomes rather than activity. A Dot completing more steps is not automatically productive, and an agent visit is not automatically valuable demand. Track completed work, human review time, errors, reversals and commercial results. Dots are an important distribution and operations shift, but the first rollout is limited and the evidence should decide where persistent autonomy earns wider access.
News date: 29 September 2026. Editorial review: 1 October 2026.