Most AI assistants wait for you to ask them something. Dots don’t. OpenAI announced dots as a new category of always-on agent, powered by GPT-6 Astra, designed to run in the background, learn your standards over time, and bring you finished work rather than a draft you still have to wrestle into shape. That’s a meaningful shift from the chat paradigm that has defined most AI tooling since 2022.
The concept of “agentic AI” has been floating around for a while. Competitors like Anthropic with Claude’s computer use, Google with Project Mariner, and a wave of startups including Cognition (Devin), Lindy, and Cohere have all been pushing in this direction. But OpenAI is going further with dots by baking in persistent memory, a dedicated cloud computer for each agent, and direct integrations with over 4,000 apps through a plugin ecosystem. This isn’t a feature addition to ChatGPT. It’s a new product layer sitting on top of it.
What dots actually does
Each dot gets its own cloud computer, its own browser, and the ability to connect to apps you authorize. It can run multiple projects in parallel without you managing the threads. A developer’s dot can watch customer feedback, scope bug fixes, write and test code, and deliver pull requests with video walkthroughs attached. A content creator’s dot can process interview transcripts, cut clips, write show notes, and draft social posts, then carry edits across every asset once you approve them.
The proactive research feature is worth highlighting. When you’re not actively working with your dot, it continues doing read-only background research using connected apps. It cannot send messages or make changes during this mode, which is a sensible guardrail. If the monitoring system flags something concerning, it can pause or stop the dot entirely.
Dots also work across platforms. You can message your dot in ChatGPT on desktop, web, and mobile, or in Slack and Teams, with SMS support coming later. Context carries across every channel, so switching from ChatGPT to Slack mid-project doesn’t break the thread.
Control and privacy
OpenAI has built a tiered permission system into dots. You choose which apps can be accessed, set custom rules for when the dot should act independently versus ask for approval, and can review all activity including background work through an Activity View. Your local computer stays separate from the dot’s cloud environment unless you explicitly connect it.
On enterprise and business plans, conversation data and dot activity are not used to train models by default. Personal plan users can opt out of that too. OpenAI says it does not train directly on proactive research outputs or the dot’s private notes.
Availability and pricing
Dots are rolling out now across Pro, Business Premium, and Enterprise plans in eligible markets. OpenAI plans to expand access soon. There’s also a preview of “specialist dots” designed for organizations, with their own identity for access management, IT-provisioned hardware, and deep integrations into company systems of record. A partnership with Microsoft brings specialist dots into Agent 365.
- Powered by GPT-6 Astra with persistent memory and learning from feedback
- Dedicated cloud computer per dot, with optional connection to your local machine
- Connects to 4,000+ apps via plugins
- Available in ChatGPT, Slack, and Teams
- Custom rules, read-only proactive research mode, and real-time monitoring
- Available on Pro, Business Premium, and Enterprise plans
Why this matters
The broader bet OpenAI is making here is that the value of AI shifts from “good answers” to “work done.” That’s a harder product to build and a harder product to trust. But if dots delivers even half of what’s described, it puts real pressure on pure-play agent startups who don’t have GPT-6 underneath them or ChatGPT’s distribution. And it gives enterprise buyers a reason to consolidate on OpenAI rather than stitching together separate tools for automation, assistants, and workflow. So the question isn’t really whether dots is impressive. It’s whether users will actually hand over the wheel.
