OpenAI Dots: Inside OpenAI's New Always-On AI Agent

Photo by Growtika on Unsplash

OpenAI just launched an agent that works while you're not looking

At DevDay 2026 in San Francisco on September 29, OpenAI unveiled dots, a new class of "always-on agents" that run continuously in the cloud and handle tasks on a user's behalf, whether or not that user is actually at their computer. It was the headline item among more than 20 announcements at the event, and it's the clearest sign yet that OpenAI wants ChatGPT to stop being something you open and start being something that's already working when you check in.

If you've used an AI agent before, you know the usual pattern: you give it a task, it runs for a few minutes or hours, then it stops and waits for you. Dots are built to skip that last part. OpenAI says each dot gets its own cloud compute, learns a user's priorities over time, and keeps working in the background, picking up new tasks, doing research, and flagging things that need a decision.

What a dot actually does

OpenAI demoed a handful of concrete use cases rather than just talking in the abstract, which is worth noting because vague "AI agent" demos have gotten pretty tiresome. The examples included a dot that noticed a bug mentioned in a Slack thread and went and investigated it on its own, one that built a working app from a set of design files, and one that scanned for invoices that hadn't been created yet and drafted them for a human to approve. Between assigned tasks, dots are also meant to do proactive research using read-only tool connections, so they're not fully idle even when nobody has given them a new job.

Setup happens through the ChatGPT desktop app, but once a dot exists you can check in on it and give feedback from your phone. OpenAI says dots can reach into more than 4,000 connected applications through its plugin ecosystem, and a user can grant one permission to access their local machine when a task calls for it. One dot comes included with a Pro or Business Premium subscription, and OpenAI has said additional dots will be purchasable later, though pricing for that hasn't been announced.

It's easy to read dots as a direct answer to Meta's Muse, the avatar-based agent Meta launched earlier this year, and several outlets covering DevDay framed it exactly that way. The framing isn't wrong, but it undersells what's actually different here. Muse leans into a visible, persona-driven interface. Dots are closer to a standing employee: something with a scope of responsibility that keeps operating whether or not you're watching, and that's a meaningfully different bet on what people want from an AI agent.

The rest of DevDay, briefly

Dots got most of the headlines, but OpenAI paired it with a real model update. GPT-6.1 Sol is pitched as bringing "near-Astra intelligence" (Astra being OpenAI's higher-end model) to the rest of the lineup at roughly a fifth of the previous input and output token cost, and it rolled out across the API, Plus, Pro, Business, Enterprise, and Edu tiers on the same day. OpenAI also introduced an "Ultrafast" speed tier that claims up to 8x faster generation in Codex and 6x faster through the API, available to Pro 500 and Enterprise customers.

On the developer side, the Codex CLI picked up voice-controlled task steering and a redesigned multi-task view, there's a new Code Review feature for GitHub and GitLab built into the desktop app, and the Agents API now supports computer use alongside multi-agent workflows. OpenAI also announced a Decisions API aimed at real-time content classification and request routing, which is a more infrastructure-flavored release than the consumer-facing agent news but matters a lot if you're building on top of the platform.

There's a collaboration push too: a new ChatGPT Space for team workspaces, shared documents called Pages that support real-time multi-user editing alongside an AI agent, and the ability to mention @ChatGPT directly inside Slack or Microsoft Teams. OpenAI says ChatGPT now has 1.2 billion weekly users, a number that gives some sense of why the company keeps pushing further into workplace tools rather than just chat.

Separately from the product news, CNBC and other outlets reported OpenAI is in talks for a new funding round around the same timeframe as DevDay, though terms weren't confirmed on stage and OpenAI hasn't detailed the round publicly.

Why this matters if you build with AI

For developers, the practical stuff to pay attention to is GPT-6.1 Sol's pricing, since a large cut in token costs for near-top-tier intelligence changes the math on what's worth automating, and the expanded Agents API, since computer use plus multi-agent support means you can build closer to what dots themselves are doing rather than waiting for OpenAI to ship the packaged version.

For everyone else, dots are worth watching less for the specific feature set (which will change) and more for what it signals. OpenAI, and its competitors after it, are converging on the idea that the next version of an AI assistant isn't a chat window you type into, but a background process with standing permissions and a memory of what you care about. That raises real questions about oversight, since a system that's aware of your priorities and can act on tools without a fresh prompt each time needs stronger guardrails than something that only reacts to explicit instructions.

It's also worth being a little skeptical of the demo-to-reality gap here. Investigating a Slack bug report and drafting a missing invoice are good demo material precisely because they're bounded, well-scoped tasks. Whether dots hold up on messier, ambiguous work that doesn't fit neatly into a five-minute demo is the thing to actually watch over the next few months, not the announcement itself.

Key takeaways

Dots are OpenAI's new always-on AI agents, announced at DevDay 2026 on September 29, that run continuously in the cloud rather than stopping after a single task. They connect to more than 4,000 applications, can access a local computer with permission, and ship with Pro and Business Premium plans, with extra dots to be sold separately later. The same event brought GPT-6.1 Sol, a cheaper near-top-tier model, an Ultrafast speed tier, and a set of developer-facing API updates including computer use and multi-agent support in the Agents API.

FAQ

Is dots available now? Yes, for Pro and Business Premium users in select markets, with beta access for Enterprise and Education customers as of the DevDay announcement.

Do I need the desktop app to use it? You need the ChatGPT desktop app to create and configure a dot. After that, you can monitor and interact with it from mobile.

How is this different from a regular ChatGPT agent task? A standard agent task runs once and stops. A dot keeps running, does proactive research between assigned tasks, and is meant to build up an understanding of a user's priorities over time.

Does this replace human oversight? No. OpenAI's own demos still show dots flagging things like invoices for human approval rather than acting fully autonomously on financial or high-stakes actions.