OpenAI has moved its flagship product from answering questions to working in the background. At DevDay 2026 in San Francisco on September 29, the company introduced dots, always-on AI agents that run on GPT-6 Astra, each with its own dedicated cloud computer and browser, and that keep working on a user's goals after the ChatGPT window is closed.
The launch is OpenAI's clearest bet yet on persistent agentic AI: software that takes on a goal, works on it for hours or days, and checks in only when it needs a decision or has something to show.
What a Dot Actually Does
A dot is a long-running assistant, not a chat session. According to OpenAI's announcement and launch coverage, each dot:
- Runs on GPT-6 Astra, the model OpenAI released on September 3
- Works from an isolated cloud computer and browser, separate from the user's own machine unless the user chooses to connect it
- Connects to more than 4,000 apps through plugins and can work on several projects in parallel
- Can be messaged or called in ChatGPT on desktop, web and mobile, and messaged in Slack and Microsoft Teams, with SMS-style messaging on a waitlist
OpenAI's examples are deliberately ordinary office work: watching an inbox, investigating a bug, and assembling a report from several apps. When a dot has no assigned task, it does what OpenAI calls proactive research in read-only mode, looking for ways to help without changing anything.
Pricing and Availability
The first dot comes at no extra cost on the Pro and Business Premium ChatGPT plans, and conversations with a dot do not count toward ChatGPT usage limits. Business Premium customers get access in all supported ChatGPT regions. Pro subscribers in the European Economic Area, Switzerland and the UK are excluded for now. Enterprise, Edu and Healthcare workspaces can enable a beta through their administrators.
At launch each user gets one dot. OpenAI has said multiple agents per user, and ways to scale an agent's speed or workload, will come later.
For businesses, OpenAI is previewing specialist dots: agents with their own identity, credentials and access to company systems for set roles such as procurement, invoicing, email marketing, customer support and contracting. OpenAI is also working with Microsoft to bring them into Agent 365, Microsoft's management layer for enterprise agents.
Guardrails Built Into the Design
An agent that acts while its owner is asleep needs firm limits, and OpenAI put its safety controls at the centre of the launch. Users decide what a dot may do on its own, what needs approval, and what it must never do, using Custom Rules to permit, block or gate specific tasks.
Other safeguards include:
- Read-only background access for proactive research
- Auto-review of actions that affect accounts or share information
- Mandatory approval for sensitive steps; The Next Web reports that a dot can prepare a password change but cannot complete it alone
- A monitoring system that can pause or stop an agent when it detects a safety concern
The timing makes these controls matter even more. A day before DevDay, OpenAI pulled its planned GPT-6.1 Astra flagship after it failed internal safety tests. Critics, including Tech Startups, noted that the company then shipped always-on agents anyway, built on the earlier Astra model.
Why It Matters
Dots move ChatGPT into the same persistent-agent space that has filled up quickly through 2026. Meta expanded its Muse personal agent for small businesses on the same day, and Microsoft rebuilt Copilot around an always-on agent just days earlier. With ChatGPT's very large user base, OpenAI can put a 24/7 agent in front of more people than any rival, so far only on its higher-priced tiers.
For enterprises, three things stand out:
- Agents are becoming identities. Specialist dots with their own credentials will need the same governance as human employees: access reviews, audit logs and offboarding.
- Isolation is the default. Running each agent on its own cloud machine, instead of the user's laptop, limits the damage a confused or manipulated agent can do.
- Approval design is the real product. How well Custom Rules match an organisation's risk tolerance will decide whether dots are trusted with real work.
DevDay also brought GPT-6.1 Sol, which OpenAI says nearly matches GPT-6 Astra on coding and computer use at about one-fifth of Astra's token price. Cheaper models of that kind are what make round-the-clock agents affordable to run at scale.
What to Watch Next
The key questions are practical. Can dots handle multi-day tasks without drifting off goal? Will the approval prompts be sensible or exhausting? And how will regulators respond? The US Federal Trade Commission has reportedly opened a broad inquiry into whether agent incidents and safety claims at frontier labs could breach consumer-protection law. The European rollout gap for Pro users also suggests OpenAI is still working through compliance questions in stricter jurisdictions.
For now, dots are the clearest sign yet that the chatbot era is giving way to delegated, always-on AI work, and that the industry's competition is shifting from model benchmarks to trust, control and integration.
