OpenAI used its DevDay keynote on Tuesday, September 29, to introduce Dots, a new class of “always-on” AI agents the company says are built to run tasks independently in the background rather than wait for a prompt. It was the headline reveal in a event that also included an upgraded GPT-6.1 model, a shared workspace called ChatGPT Space, and a new API aimed at letting companies hand off small, repetitive decisions to AI — a lineup that, taken together, is OpenAI’s clearest attempt yet to move ChatGPT from something people chat with into something that works on its own.
What Dots Actually Does
Dots is built on OpenAI’s Astra model and described by the company as an agent designed to “handle everything,” operating continuously across devices and interfaces rather than running a single task and stopping. Users get a primary Dot they can name and personalize, and can spin up specialist variants with their own credentials, tools, and areas of responsibility — OpenAI’s pitch is that several Dots can eventually work together on a person’s behalf rather than one assistant trying to do everything at once. The company’s example use cases lean practical rather than flashy: a developer’s Dot monitoring customer feedback channels and shipping small bug fixes on its own, or a researcher’s Dot rerunning an analysis overnight and flagging an unexpected result by morning.
Access isn’t limited to the ChatGPT app. Dots can be reached through Slack and Microsoft Teams, with SMS access planned, and OpenAI has already lined up a partnership with Microsoft to route Dots through Agent 365, Microsoft’s security and governance layer for AI agents operating inside a company’s existing tools. That enterprise integration matters more than it might sound: an agent that can act inside Slack or Teams needs the kind of permissioning and audit trail that IT departments actually require before they’ll let it near real work.
A Bubbly Face for a Serious Push Into Autonomy
Visually, Dots leans cartoonish on purpose — a cluster of floating, animated dots stands in for the agent’s “face,” a design choice OpenAI has framed as making an increasingly autonomous piece of software feel more approachable rather than more alarming. Multiple outlets covering the launch, including TechCrunch, noted the resemblance to Meta’s Muse, the bubbly agent avatar Meta shipped earlier this year, and Axios described Dots outright as OpenAI’s answer to it. The comparison points to a broader pattern: as AI agents take on more independent, always-running responsibility, the companies building them seem to agree that a friendlier, less utilitarian interface is part of getting people comfortable handing over the keys.
Dots is rolling out starting this week to ChatGPT Pro and Business Premium subscribers in eligible markets, accessible through both ChatGPT and Codex. OpenAI hasn’t disclosed separate pricing for the feature itself, and there’s no confirmed timeline yet for a wider rollout to lower-tier plans.
The Rest of the DevDay Lineup
Dots wasn’t the only launch. OpenAI also introduced GPT-6.1 Sol, a version of GPT-6 tuned to approach the performance of the flagship Astra model on coding and computer-use tasks at a significantly lower cost — effectively a cheaper, faster option for developers who don’t need the top-tier model for every request. Alongside it came ChatGPT Space, a persistent shared workspace where teams, ChatGPT, and Dots can collaborate on a project together, blending pieces of what Slack, Notion, and Google Drive each do with AI agents built directly into the workflow rather than bolted on as a plugin.
OpenAI also unveiled a Decisions API, aimed at companies that want to automate narrow, repetitive calls — approve or reject a refund, flag or clear a transaction — by defining a fixed set of possible outcomes and letting the model choose among them, rather than generating open-ended text. And Codex, OpenAI’s coding agent, is moving further into the cloud: developers can now run it remotely and manage it from any device, with shared development environments meant to let human engineers and the agent work in the same space. CEO Sam Altman used the keynote to tease two things still to come — dedicated AI hardware and a separate safety platform for agents — without giving specifics on either.
An IPO Still Waiting in the Wings
The product news came with a smaller but notable business update: Bloomberg reported that Altman told attendees OpenAI’s investors have been patient about a public listing while the company keeps its focus on safety rather than IPO timing. It’s a continuation of a theme that’s been building all year — OpenAI has repeatedly downplayed near-term IPO plans even as its valuation and product footprint keep expanding, and DevDay’s focus on always-on agents, enterprise partnerships like the Microsoft Agent 365 integration, and infrastructure like Codex in the Cloud all read as the kind of durable, revenue-generating groundwork a company lays before it eventually does go public, whenever that turns out to be.
Why This Round of Announcements Is Different
What separates this DevDay from previous ones is the shift in framing: OpenAI isn’t just shipping a smarter model, it’s shipping software designed to be left running. A chatbot waits for you to type; an “always-on” agent doesn’t. That distinction is where the real competitive fight for 2026 seems to be heading, with Meta’s Muse, Google’s various Gemini agent tools, and now Dots all racing to become the thing that quietly handles tasks in the background rather than the thing you open a tab to talk to. Whether users actually want that much autonomy handed to a bubbly cluster of animated dots — and whether enterprises trust it with real credentials and real decisions — is the question OpenAI is now betting the next phase of ChatGPT on.