OpenAI Dots (announced September 29, 2026 at DevDay in San Francisco) are always-on AI agents in ChatGPT powered by GPT-6 Astra. Unlike chat models that wait for prompts, Dots take on ongoing responsibilities, execute multi-step workflows on dedicated cloud computers, remember context, and run across connected apps with human approval guardrails.
1. The Big Idea: From Answering Prompts to Owning Responsibilities
The fundamental interface of modern computing is undergoing its most radical transformation since the browser. For the past three years, corporate adoption of artificial intelligence was dominated by conversational turn-taking: a human types a prompt, an LLM outputs text, and the human manually copies that text into their software workflows.
“AI is moving from answering questions to owning responsibilities. Software used to wait for humans; agentic software works continuously between human decisions.”
To understand this shift, enterprise technology leaders must distinguish between the four developmental stages of corporate AI tools:
2. What Exactly Are OpenAI Dots?
OpenAI explicitly defines Dots as always-on agents inside ChatGPT designed to take on ongoing responsibilities and continue making progress on complex objectives between user conversations.
CRITICAL TECHNICAL CORRECTION: Dots are NOT a new AI model. They represent a specialized agent product experience built within ChatGPT. The underlying intelligence is powered by OpenAI's flagship GPT-6 Astra foundation model.
4. Motion Graphic: Live Agent Workflow Simulation
Experience how a Dot agent executes a multi-step background objective continuously inside ChatGPT:
5. When and Where Was Dots Announced?
OpenAI Dots were unveiled during the opening keynote of OpenAI DevDay 2026 on September 29, 2026 at Fort Mason, San Francisco.
DevDay 2026 featured over 20 major technological announcements spanning ChatGPT capabilities, Codex engine updates, API access tiers, and new agentic developer frameworks.
6. How Dots Differ from Traditional ChatGPT
The architectural differences between standard ChatGPT sessions and OpenAI Dots lie in state persistence, compute isolation, and continuous execution context.
| Dimension | Traditional ChatGPT | OpenAI Dot (Always-On Agent) |
|---|---|---|
| Interaction Model | Synchronous prompt → instant reply | Asynchronous responsibility → continuous progress |
| Task Model | Single question or isolated transaction | Multi-step goal spanning days/weeks |
| Persistence & State | Clears state when session or window closes | Persistent memory across conversations |
| Computer Infrastructure | Stateless web runtime sandbox | Dedicated cloud computer instance |
| Application Connectivity | Manual paste or basic browser search | Direct integration with user-authorized apps |
| Execution Engine | Waits for user input at every turn | Executes background loops until milestone |
| Human Oversight | Human inspects every single prompt output | Human reviews high-impact decision checkpoints |
7. The Six Core Features of OpenAI Dots
1. Always-On Execution
Dots do not terminate when you close ChatGPT. They remain active in the cloud, monitoring events and completing steps toward your objectives.
2. Dedicated Cloud Computer
Each Dot operates within its own cloud computer environment, enabling web navigation, file manipulation, data transformation, and CLI script execution.
3. Connected App Ecosystem
Dots connect across authorized enterprise software—CRMs, LMS platforms, email, calendars, cloud storage, and analytics systems.
4. Persistent Context Memory
Dots maintain historical context, corporate guidelines, project constraints, and historical decisions without losing focus across sessions.
5. Goal-Based Delegation
Users define macro responsibilities (e.g., "Summarize weekly admissions data") rather than micro-managing step-by-step commands.
6. Human Review Boundaries
When an agent hits policy boundaries or outbound actions (e.g. sending emails, making financial edits), it pauses and requests explicit human approval.
8. Under the Hood: What an Always-On Agent Actually Requires
Building a reliable always-on agent system requires far more than wrapping an LLM call in a loop. It demands an enterprise-grade 8-layer architecture spanning compute, memory, permissions, and verification engines.
Conceptual representation of the internal architectural pattern required for continuous autonomous execution inside ChatGPT.
9. The Most Important Paradigm Shift in Enterprise Software
The true significance of OpenAI Dots is not that a chatbot can now browse the web longer. It is that the primary role of business software is shifting from tools humans operate to systems humans supervise.
10. Real Business Use Cases for Always-On Agents
While OpenAI Dots will roll out capabilities incrementally, enterprise systems architects are already designing operational workflows around background agent patterns:
Market Change Monitor
Goal: Continuously monitor competitor announcements, pricing updates, and industry news every morning, digest key signals, draft an executive summary, and submit for review.
Lead Lifecycle Reconciler
Goal: Monitor incoming web inquiries, cross-reference data across CRM and LinkedIn, flag missing contact details, prepare customized response drafts, and alert account executives.
Automated Report Synthesizer
Goal: Query operational analytics databases weekly, calculate variance metrics, draft an executive memo highlighting anomalies, and schedule a leadership review.
Knowledge Base Auditor
Goal: Track recurring customer support ticket themes, identify knowledge gaps in product documentation, write updated help draft pages, and ping product managers for approval.
11. What Could Always-On AI Mean for Universities & LMS Platforms?
As higher education specialists who design custom Moodle LMS architectures and university digital ecosystems, Filari Agency views always-on agentic technology as a catalyst for campus transformation.
Note: These represent potential architectural use cases for enterprise university digital transformation, not claims that OpenAI Dots natively provide pre-built integrations for every specific university software platform out of the box.
12. Architectural Breakdown: Chatbot vs Agent vs Always-On Agent
Chatbot Loop
↓
LLM Inference
↓
Text Output
↓
[Session Terminates]
Agentic Task Loop
↓
Plan Steps
↓
Execute API Tools
↓
Verify Results
↓
[Return Final Output]
Always-On Loop
↓
Observe State & Events
↓
Plan & Execute in Cloud
↓
Escalate at Decision Boundary
↓
[State Saved · Loop Continues]
13. Permissions & Governance: The Hard Problem Is No Longer Intelligence, It Is Authority
When an AI system is granted the capability to execute actions autonomously on cloud computers and across corporate databases, the primary bottleneck shifts from intelligence to organizational authority and permission boundaries.
OpenAI's official documentation highlights that enterprise workspace administrators maintain granular controls over Dots access, tool usage, and connected application boundaries.
| Permission Scope | Autonomous Action | Governance Control | Status State |
|---|---|---|---|
| READ | Read documents, query APIs, fetch emails, search database | User/Admin authorized OAuth tokens | [ Unrestricted Read ] |
| WRITE | Draft reports, create database rows, update CRM records | Workspace Policy Engine rules | [ Sandbox Staged ] |
| EXECUTE | Run scripts on cloud computer, navigate external websites | Cloud Sandbox Isolation | [ Isolated Compute ] |
| COMMUNICATE | Send emails, post Slack messages, contact leads | Mandatory Human Approval Gate | [ Human Approval Required ] |
| APPROVE | Financial payouts, policy modifications, deletion of data | Blocked by Default Policy | [ Admin Blocked ] |
14. Enterprise Security & Auditability Stack
Deploying always-on agents requires strict identity verification, role-based access control (RBAC), immutable logging, and clear escalation protocols.
↓
ROLE-BASED ACCESS CONTROL (RBAC)
↓
AGENT EXECUTION ENGINE (GPT-6 Astra)
↓
POLICY ENGINE & GATEKEEPER (Evaluates least-privilege rules)
↓
IMMUTABLE AUDIT LOG & COMPLIANCE LEDGER
↓
HUMAN REVIEW ESCALATION PANEL
15. Availability & Regional Access Grid
OpenAI is rolling out Dots in phased tiers across ChatGPT plans and geographic jurisdictions.
| Plan Tier | Availability Status | Regional / Admin Conditions |
|---|---|---|
| ChatGPT Pro | Rolling out to supported markets | EXCLUDES: European Economic Area (EEA), Switzerland, and the United Kingdom at launch. |
| Business Premium | Available | Supported across active ChatGPT business regions. |
| Enterprise | Beta Access | Off by default. Requires workspace administrator enablement in settings. |
| Edu (Universities) | Beta Access | Subject to enterprise workspace admin controls and institutional availability. |
| Age Requirement | Mandatory | Eligible users must be 18 years or older. |
Source: OpenAI Help Center — Getting started with your dot. Last verified: October 2026.
16. What OpenAI Dots Are NOT: Myths vs. Facts
Dots are persistent agents engineered around ongoing responsibilities, cloud computing, and asynchronous progress.
Dots are an agent product experience inside ChatGPT, powered by the existing GPT-6 Astra model.
Access is strictly gated by connected app OAuth permissions, user choices, and enterprise workspace policies.
Important actions, outbound communications, and critical decision boundaries require explicit human review.
17. From Chat Interface to Workflow Architecture
As AI tools mature, the user interface becomes less important while the underlying workflow integration becomes decisive.
OLD CHAT MODEL
Human → Manual Prompt → AI Response → Human executes every task manually across software apps.
NEW AGENTIC MODEL
Human defines Responsibility → AI Agent Plans → Executes on Cloud Compute → Connects to Apps → Human Supervises Output.
18. What This Means for Software & SaaS Architecture
SaaS products built over the past decade were designed around human point-and-click UX. In the agentic era, software companies must design agent-ready APIs, structured event streams, permission layers, and audit trails.
Software is transforming from software humans operate to systems humans supervise.
19. Filari’s Point of View: Design Around Responsibility, Not Conversation
At Filari Agency (filari.agency), our work with higher education institutions, Moodle LMS platforms, EdTech ecosystems, and enterprise automation leads us to a clear conclusion:
“The biggest strategic opportunity is not adding another chatbot widget to your website or learning platform. It is redesigning organizational workflows so AI can continuously handle the work that sits between human decisions.”
Whether building adaptive learning systems for universities, streamlining student enrollment funnels, or integrating enterprise CRMs, the winning pattern is clear: start with the workflow, define the responsibility, enforce permission boundaries, and keep humans in control of key decisions.
20. The Evolution Timeline of AI Work (2023–2026+)
21. The Executive Questions Every Leadership Team Must Ask
If AI can work continuously in the background, enterprise leadership must address six fundamental questions:
- 1. Who defines the agent's ongoing responsibility?
- 2. Who grants the agent tool and data access permissions?
- 3. Who audits the agent's continuous actions?
- 4. Who approves high-risk or financial decisions?
- 5. Who ultimately owns the operational outcome?
- 6. What happens when the agent makes an error?
22. Frequently Asked Questions (AEO & Schema Ready)
OpenAI Dots are always-on AI agents inside ChatGPT that take on ongoing responsibilities and continue making progress on multi-step workflows between user conversations. They operate on dedicated cloud computer environments and connect across user-authorized applications.
OpenAI Dots were officially announced on September 29, 2026, during the opening keynote of OpenAI DevDay 2026 at Fort Mason in San Francisco.
No. OpenAI Dots are an agentic product experience inside ChatGPT, powered by the underlying GPT-6 Astra foundation model. Dots are not a new standalone model.
OpenAI Dots are powered by OpenAI's GPT-6 Astra foundation model, engineered specifically for multi-step reasoning, tool usage, and continuous workflow execution.
Traditional ChatGPT operates on a synchronous prompt-and-response model that ends when the conversation stops. OpenAI Dots maintain persistent state, operate on their own cloud computers, and execute ongoing responsibilities asynchronously between user logins.
Yes. Because Dots are always-on agents running on cloud computers, they can continue background monitoring, data analysis, and task execution according to your defined schedule and guardrails while you are offline.
Yes. OpenAI Dots execute work inside a secure, dedicated cloud computer environment capable of handling browsers, files, APIs, and command tools.
Yes. Dots can interact with user-connected applications such as CRMs, email, calendars, cloud storage, project tools, and analytics platforms, strictly governed by user permissions and enterprise admin policies.
No. Initial rollout for ChatGPT Pro users excludes the European Economic Area (EEA), Switzerland, and the United Kingdom. Business Premium is available in supported regions, while Enterprise and Edu access is offered via beta requiring workspace admin enablement.
Yes, in beta. Access for Enterprise and Edu accounts is off by default and requires workspace administrator enablement through ChatGPT workspace controls.
Yes. Enterprise workspace administrators can configure feature availability, connected application permissions, data retention policies, and action boundaries across their organization.
No. OpenAI Dots incorporate human-in-the-loop governance. When a Dot encounters a critical decision boundary, policy threshold, or external communication step, it pauses and escalates the action for human review and approval.
Key considerations include role-based access control (RBAC), least-privilege scoping, explicit tool permissions, comprehensive audit logs, data isolation, and policy enforcement engines.
In higher education, always-on agents can continuously reconcile admissions leads, monitor student engagement in LMS platforms (e.g. Moodle), assist faculty with course admin, and flag at-risk students for intervention.
SaaS applications are shifting from software humans manually operate toward underlying systems that autonomous agent layers execute while humans supervise decisions.