At OpenAI DevDay 2026, the most significant announcement wasn’t a new model — it was a new class of product. Dots are personal AI agents built on the Astra model that run persistently inside your existing tools, not just when you open a chat window.
What Dots Are
A Dot is a named, customizable agent that you configure once and grant ongoing access to specific integrations: Slack, Microsoft Teams, email, and via Computer Use, the broader desktop environment. Unlike session-based AI assistants that reset between conversations, Dots maintain context across interactions and can initiate work without being prompted.
The setup is deliberate: you name the Dot, choose an avatar, and define its access scope. OpenAI is framing this as granting varying levels of responsibility — a Dot with Slack access can read and post messages; a Dot with Computer Use can navigate applications on your behalf.
The ChatGPT Space Integration
Dots connect to a new ChatGPT Space layer designed for team collaboration. Multiple team members can share a Space, and the Dot builds artifacts — documents, analysis outputs, task lists — that persist and update as the shared context evolves. This is a direct move at Microsoft Copilot’s territory: an AI that doesn’t just answer questions inside a chat interface but holds and updates shared work products.
Voice interactions are built in, enabling conversational task delegation rather than typed prompts. This matters for the use cases where AI assistance competes with a human assistant: “Add a follow-up item from today’s Slack thread to the project doc” is faster spoken than typed.
What Computer Use Adds
Dots with Computer Use access can navigate legacy applications and perform multi-step actions across the desktop environment. The positioning at DevDay was around complex tasks: a Dot that monitors an inbox, extracts action items from emails, cross-references a CRM, and updates a spreadsheet — without requiring API integration at each step.
This is the practical bridge between “AI-assisted work” and “AI-automated work.” Computer Use as a raw capability has existed for months; Dots packages it with persistent identity, defined scope, and a user-facing permission model that makes it deployable without engineering overhead.
The Astra Model Foundation
Running Dots on Astra (rather than a cheaper model) is a deliberate product choice. Astra’s capability tier is what enables the multi-step reasoning, context retention, and tool orchestration that persistent agents require. The GPT-6.1 Sol launch at the same event offers near-Astra capability at a fifth of the cost — but Dots specifically requires Astra’s reliability for the ongoing-agent use case.
What This Means for Agentic AI Stacks
Dots is not a developer platform — it’s a consumer and enterprise product with a configuration UX. But it operationalizes several patterns that the agentic AI community has been building toward in code:
- Persistent agent identity — a named, scoped agent rather than stateless API calls
- Cross-tool integration without custom APIs — Computer Use as the fallback for anything without a native integration
- Human-authorized workflows — explicit scope grants at setup rather than ad-hoc permission requests
- Shared context layers — artifacts that multiple agents and humans read and write
The pattern Dots establishes — always-on, scoped, integrated — is the architecture that enterprise agentic deployments are converging on regardless of which model or platform they run on.
The So What
Dots represents the productization of agentic patterns that have been engineering-only for the past year. When OpenAI ships a consumer interface for persistent, tool-integrated agents, it sets the expectation for what “AI agent” means to non-developers. That shapes what enterprise buyers demand, what competing platforms build toward, and what the next generation of developer tooling assumes as baseline.
For teams building agentic systems: the Dots model — persistent identity, defined scope, Computer Use as universal integration layer — is increasingly the target architecture to reverse-engineer from, whether you’re building on OpenAI or not.
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →