ChatGPT Dots Explained: How OpenAI's Always-On Agents Work

On September 29, 2026, OpenAI used its Dev Day to introduce ChatGPT Dots: remarkably capable, always-on agents built to handle everything. The idea is simple to state and demanding to run well. Instead of chatting with an assistant that forgets the job when you close the tab, you hand ongoing work to a Dot that keeps going in the background on its own cloud computer, checks back when it needs you, and respects approval gates you set in advance.

This guide explains what Dots are, how they operate across your apps, how permissions and supervision work, who can use them today, and where the honest unknowns sit. If your interest is practical deployment rather than the concept, pair this with an AI internal assistant scoped to one workflow first.

Key takeaways

• ChatGPT Dots launched September 29, 2026 at OpenAI Dev Day, powered by the GPT-6 Astra model.
• Each Dot runs on its own isolated cloud computer and browser, separate from your laptop, and connects to more than 4,000 apps.
• You reach Dots through ChatGPT, mobile apps, Slack, and Microsoft Teams, with voice calls supported and SMS planned.
• Background research runs read-only; actions can be allowed, blocked, or require approval, supervised through an Activity View.
• Rollout covers ChatGPT Pro and Business Premium in eligible markets, with enterprise, education, and healthcare as an admin-enabled beta.
• Specialist Dots for procurement, invoicing, support, and contracts are in testing, with Microsoft Agent 365 governance on the roadmap.

What a Dot actually is

A Dot is a persistent agent with three parts: a job it owns, a computer it runs on, and a set of connections into your tools. The computer matters most. Each Dot gets its own cloud computer and browser, an isolated virtual machine rather than a tab on your laptop. That isolation means a Dot can browse, fill forms, and operate web apps without disturbing your machine or depending on it staying awake.

The second part is reach. Dots connect to more than 4,000 apps through OpenAI's plugin ecosystem, and they work across interfaces rather than living inside a single window. You can start something in ChatGPT, check it from mobile, discuss it in Slack or Teams, or handle it by voice call. SMS support is planned but not yet available, so teams that coordinate over text should treat that channel as roadmap rather than reality.

The third part is continuity. Dots keep working after you leave, then return for input when an approval is required. The practical effect is delegation with checkpoints rather than fire-and-forget automation: you assign, the Dot progresses, you approve or redirect at defined gates.

How Dots do proactive work

OpenAI draws a clear line between looking and acting. Background research runs read-only, which means a Dot can gather, compare, and summarize without changing anything in your systems. Actions are a separate decision: each category of action can be allowed outright, blocked entirely, or set to require your approval first. That three-way setting is the single most important configuration in the product, and owners should review it per Dot rather than accepting defaults.

Supervision happens through the Activity View, where you watch what a Dot is doing and redirect it mid-task when it drifts. Redirection beats restarting: correcting course early keeps context intact and teaches you what your briefs leave ambiguous. Two further guardrails are worth knowing. Sensitive tasks such as changing a password always stay with the user rather than being delegated, and credentials can be used by a Dot without exposing passwords to the model. Dots can also be stopped automatically if systems detect malicious instructions, which is a backstop against hijacked tasks, not a substitute for scoped permissions.

For background on why these gates matter, our AI agent governance notes walk through approval tiers and review habits in plain terms.

Availability, plans, and regional limits

Dots are rolling out to ChatGPT Pro and Business Premium in eligible markets. The first Dot is included, with extended limits during the first month, and Dot usage does not count toward plan limits initially. That introductory posture is generous and temporary, so price your pilot on the assumption that metering normalizes: measure consumption per job now so later limits do not surprise you.

Enterprise, education, and healthcare access runs as a beta that an administrator must enable, and it ships off by default. That default is a sensible signal. Organization-wide agents should be an explicit decision with named owners, not something that activates quietly. Two regional caveats apply at launch: Pro is initially unavailable in the EEA, Switzerland, and the UK, while Business Premium is available across all supported regions. If your team spans those areas, confirm eligibility seat by seat before promising a rollout date.

Specialist Dots and the enterprise story

The most business-relevant part of the announcement is still in testing: specialist Dots for procurement, invoice processing, customer support, and commercial contracts. Each specialist gets its own identity with access to selected company systems, which is a meaningful step beyond a general helper with broad credentials. Scoped identities make it possible to say the invoicing Dot may read the accounting system and draft entries but may not send payments, and to enforce that technically rather than by policy memo.

OpenAI is working with Microsoft to bring these specialists into Microsoft Agent 365 security controls. Agent 365 is Microsoft's control plane for governing agents: a unified registry, per-agent identity in Microsoft Entra, lifecycle management, conditional access, threat protection, and audit. For regulated or multi-team companies, that alignment is the difference between a pilot and a deployment. Smaller firms can borrow the discipline without the platform: one identity per Dot, least-privilege access, approval gates on external actions, and logs you review weekly.

GPT-6 Astra: what is known and what is not

Dots are powered by GPT-6 Astra, OpenAI's new model. That is the extent of what launch coverage firmly establishes. Independent benchmarks, pricing details, and comparative evaluations were not published alongside the announcement, so any claim about how far ahead Astra sits should be treated as marketing until evidence appears. This is not a criticism of the model; it is the normal gap between a launch event and public measurement.

The right response is to evaluate Dots the way you would evaluate a hire during probation: on output quality, error rate, turnaround time, and supervision cost for your specific jobs. A model that excels at demos can still stumble on your data, your tools, and your edge cases. Run the same two-to-four-week scorecard you would run for any agent, and let your numbers carry more weight than launch adjectives. Our comparison of agents versus conventional software gives a framework for deciding which jobs even deserve that trial.

What Dots change about apps

One strategic read worth noting comes from TechCrunch's analysis of the launch: OpenAI is pushing ChatGPT toward an app-discovery and launch surface, a posture that challenges the traditional app-store model. If Dots become the place where people find and invoke software through conversation, distribution shifts toward whoever owns the agent layer. For a small business, the near-term implication is narrower: prefer tools with clean Dot-compatible connections and export paths, and avoid deepening dependence on any single surface until the dust settles. Delegation should increase your leverage, not your lock-in.

Who Dots suit, and who should wait

Dots suit teams that already coordinate in ChatGPT, Slack, or Teams, carry a backlog of research and cross-app errands, and can name someone to own approvals. They suit owners who will set read-only defaults first and open action permissions gradually. They do not suit teams whose core pain is inbound calls and booking, who lack anyone to review agent output, or who need SMS-based delegation today. They also do not suit organizations that cannot grant scoped credentials safely; fix identity and access hygiene before adding persistent agents.

FAQ

What are ChatGPT Dots?

ChatGPT Dots are always-on AI agents announced at OpenAI Dev Day on September 29, 2026 and powered by GPT-6 Astra. Each Dot gets its own isolated cloud computer and browser, connects to more than 4,000 apps, works in the background, and can be reached through ChatGPT, mobile, Slack, and Microsoft Teams.

How do ChatGPT Dots handle permissions and safety?

Background research runs read-only by default, while actions can be allowed, blocked, or set to require approval. An Activity View lets users watch and redirect a Dot mid-task, sensitive actions like changing a password always stay with the user, and Dots can be stopped if systems detect malicious instructions.

Who can use ChatGPT Dots right now?

Dots are rolling out to ChatGPT Pro and Business Premium in eligible markets, with the first Dot included and extended limits during the first month. Enterprise, education, and healthcare access runs as an admin-enabled beta that is off by default. Pro is initially unavailable in the EEA, Switzerland, and the UK.

What are specialist Dots for businesses?

OpenAI is testing specialist Dots for procurement, invoice processing, customer support, and commercial contracts, each with its own identity and access to selected company systems. OpenAI is working with Microsoft to bring these specialists under Microsoft Agent 365 security controls for governed enterprise use.

Next step

Pick one delegable weekly job and run it through a Dot with read-only research and approval-gated actions for two weeks, scoring hours saved against review time. If you want help choosing that first job or confirming the bottleneck is not actually phones and follow-up, book a call. To map your leaks first, run the free six-step plan: start your AI automation plan.