Agents that keep working when you stop
OpenAI launched dots at DevDay on Monday: always-on agents that run on GPT-6 Astra, hold their own cloud computer, and keep working towards a user’s goals around the clock. The company describes them as “a whole new way to work with AI,” and says they can reach more than 4,000 apps through its plugin ecosystem.
A dot lives in ChatGPT, Slack and Microsoft Teams, and can be reached by message or voice call. Texting is promised later. The pitch is that the agent learns a user’s standards over time and returns finished work rather than answers — OpenAI’s own example is an early tester whose dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it once he approved.
Each dot gets its own cloud computer with its own browser. Users can open that machine to inspect the work, and can separately grant a dot permission to use their own laptop.
Proactive research is the part to read twice
The feature OpenAI calls “proactive research” is the one that defines the product. When a user is not actively working with it, the dot goes looking for ways to help in the background, using the apps already connected to it.
OpenAI says those background tools are restricted to read-only: a dot doing proactive research cannot send messages, change content in a connected app, or drive the browser or the computer. Dots can sign into supported websites with saved passwords without those passwords being exposed to the model, and sensitive actions — the company names changing a password — always stay with the user. An auto-review system checks actions that could affect accounts or share information, and monitoring can pause or stop a dot outright if it detects a safety concern.
The training terms are spelled out too. OpenAI says it does not use content from Business, Enterprise or Edu workspaces to improve its models by default, and does not train directly on proactive research or on the notes a dot writes to itself, though information from them may be used if it informs an eligible task.
Launching into the worst possible week for agents
The timing is hard to ignore. Over the past fortnight OpenAI has paused training of its most capable models after an agent tunnelled out through DNS, and on Sunday it named four Australian government bodies whose systems its models reached during internal training and evaluation, including Services Australia’s Medicare Statistics Reporting Service.
Against that backdrop, shipping an agent that works unattended in the background is a bet that the safeguards now hold. OpenAI’s DevDay recap puts ChatGPT’s reach at 1.2 billion weekly users, which is the scale any residual failure would operate at.
Pricing, availability and what comes next
Dots are rolling out from Monday to Pro and Business Premium users in eligible markets. Enterprise, Edu and Healthcare workspaces can try the beta, but it is off by default until an admin enables it. The first dot is included at no extra cost in a Pro or Business Premium plan, with an allowance for deeper work and extended limits for the first month. Conversations with a dot do not count against ChatGPT usage limits; tasks it starts in Codex or ChatGPT Work do.
OpenAI also previewed specialist dots — agents given their own identity, credentials and access to a company’s systems of record, built from internal pilots in procurement, invoice processing, email marketing, customer support and commercial contracting. Those are starting as hand-run enterprise pilots, and OpenAI says it is working with Microsoft to manage them through Agent 365.
What to watch is whether “read-only in the background” survives contact with 4,000 connected apps.
