OpenAI Dots: Always-On AI Agents Launched at DevDay 2026

OpenAI Just Launched Dots — Always-On Agents

OpenAI announced Dots at DevDay 2026 on September 29 in San Francisco — a new class of AI agent designed to run continuously in the background, pursue user-defined goals with minimal oversight, and work across connected apps without needing a prompt for every step.

The company describes them as "remarkably capable, always-on agents built to handle everything." The pitch is a shift from the conversational AI most people know — type a question, get an answer — to persistent agents that keep working after the chat window closes.

OpenAI dots are powered by GPT-6 Astra, OpenAI's flagship model, and each agent gets its own cloud computer, browser, memory of user preferences, and access to connected applications. OpenAI says they connect to more than 4,000 apps by its count.

CEO Sam Altman announced Dots on stage at DevDay, positioning the launch as a move from answering prompts to taking on ongoing work. In his X announcement, Altman described Dots as a way to use AI around the clock.

What a Dot Actually Is

A Dot is not a chatbot. It is an agent that runs on a persistent cloud computer — a dedicated environment with its own browser and compute that stays active even when the user is not interacting with it.

That architecture matters. A chatbot is reactive: it does something when you ask. A Dot is proactive: it can continue working on a task, monitor for new information, and take steps toward a goal without waiting for the next prompt. OpenAI's announcement says each Dot learns more about a user's preferences over time as it works.

The 4,000+ app connections are the integration layer. OpenAI says Dots reach those apps through plugins, which means the agent can operate across the software a user already relies on — calendars, messaging tools, document platforms, code repositories, and the rest of the stack — rather than being confined to the ChatGPT interface.

How You Work With a Dot

OpenAI is rolling out Dots to ChatGPT Pro subscribers, who pay $100 a month, and to Business Premium users in eligible markets. The first Dot is included at no extra cost for those tiers.

Enterprise, Education, and Healthcare customers can try a beta if their workspace administrator enables it.

A user starts by setting up a primary Dot through ChatGPT on desktop or a desktop browser. From there, the Dot can be reached in several ways:

• ChatGPT — the primary interface for setup and control.
• Slack and Microsoft Teams — users can message their Dot through these organizational platforms, with the user's context shared across channels.
• iMessage and RCS — OpenAI says Pro users can join a waitlist to text with their Dot via iMessage or RCS messaging on Android, with text message support coming soon.

OpenAI says conversations with a Dot do not count toward ChatGPT usage limits. However, tasks that a Dot starts or manages in Codex or ChatGPT Work do count toward those limits. The company says plans will include an allowance for deeper work but did not specify the size of that allowance in the launch post.

Specialist Dots for Enterprise

Altman also announced "specialist Dots" on stage — agents tuned for specific enterprise tasks. The examples OpenAI gave were accounting, email marketing, and legal analysis.

Specialist Dots can be provisioned with specific identities, credentials, and tools through existing systems. That means an enterprise could deploy a Dot with access to a defined set of company resources and a defined scope of work, rather than giving a general-purpose agent the keys to everything.

OpenAI is already working with Microsoft to integrate Dots into the company's Agent 365 security controls. That integration is a signal about the enterprise angle: OpenAI is not just selling a consumer agent here; it is building toward a deployment model where the agent's access is governed by the enterprise security stack the customer already runs.

The Meta Muse Comparison

Dots are not OpenAI's first agent, and they are not the first always-on personal agent on the market.

Meta's Muse agent has been the recent benchmark for this category — it launched to significant attention and reached a broad user base, with users experimenting with a personal agent that can handle tasks across their digital life. WIRED frames Dots explicitly as OpenAI's answer to Muse.

The visual branding reinforces the comparison. Dots are depicted as cute, customizable blob characters — a bubbly, cartoonish persona that is a departure from OpenAI's usual software aesthetic. TechCrunch describes them as "little floating cartoons" and notes the branding is a makeover for OpenAI's software, part of a broader trend of companies trying to make AI feel more relatable.

The competition is not just Meta. The broader personal-agent category has been building through 2026, with early adopters experimenting with tools designed to run continuously on a user's behalf. OpenAI is now entering that market with a flagship-model-backed agent and the distribution of ChatGPT Pro.

What Dots Can Do — According to OpenAI

OpenAI's announcement and demo examples describe a range of tasks OpenAI dots can handle:

• Proactive scheduling and logistics. In one demo, a Dot noticed its user's calendar showed they were working through dinner and proactively messaged with two Grubhub food delivery options, including pricing for each. The user then confirmed their pick and when they wanted it delivered.
• Launching and building things. Another demo showed a user collaborating with a Dot to launch a new website.
• Monitoring and acting on feedback. OpenAI describes a scenario where a software developer deploys a dedicated Dot to monitor customer feedback, implement bug fixes and requested features as necessary, and present pull requests for review.
• Revising work as scope changes. OpenAI gives the example of an agent revising launch materials when a product's scope changes.

These are scenarios supplied by OpenAI, not independent evidence of reliability. OpenAI itself cautions in the launch post that Dots can make mistakes and tells users to review consequential work.

The Safety and Permission Model

OpenAI is building approval into the Dots workflow, which is the right place to start for an agent that can act across a user's apps.

The company says Dots are designed to get explicit approval before taking more sensitive actions on a user's behalf — things like installing software or changing a password. Users can enable a Custom Rules tool to set explicit boundaries they do not want agents to cross, as well as tasks that need more direct permission to complete.

For background work, OpenAI says proactive research uses read-only access when the user is not actively working with the agent. Actions that could affect accounts or share information are checked, and some sensitive actions remain with the user.

That combination — broad access plus explicit review controls — is central to the product's value proposition. An agent that can see more of a user's work is more useful, but each new permission raises the cost of a mistake. OpenAI is betting that the review gates are enough to make the trade-off worth it.

The privacy angle is worth noting. OpenAI cautions that if a user's account is set to allow model training on their data, that setting carries over to interactions with Dots. Users connecting their data and apps to a persistent agent should understand what that means for how their information is used.

The GPT-6 Astra Connection

Dots are powered by GPT-6 Astra, OpenAI's current flagship model.

That is a meaningful detail given the news cycle. OpenAI scrapped GPT-6.1 Astra — the next-generation model expected in October — on September 28 over safety concerns, one day before DevDay. GPT-6.1 Astra is not shipping. But GPT-6 Astra, the current flagship, is, and OpenAI dots are built on it.

The distinction matters for how to read the launch. Dots are not running on a model that OpenAI cancelled for safety issues. They are running on the model OpenAI is currently shipping. Whether GPT-6 Astra's own safety record — it was rated "Critical" on ExploitBench, as AIPress covered — is sufficient for a persistent agent that acts across a user's apps is a separate question that the launch does not answer.

The Bigger Picture: From Chatbot to Coworker

Dots represent a shift in how OpenAI wants people to use AI.

The ChatGPT model — the one most of the public knows — is conversational. You type, it answers. You ask, it responds. The interaction is turn-by-turn.

Dots are the next step in OpenAI's product trajectory: an agent that is assigned a responsibility and then left to pursue it. The user sets the goal, defines the boundaries, and the Dot works on it continuously. OpenAI envisions "teams of Dots working together on your behalf" over time.

Altman has a history of pushing product ambition in this direction. Before OpenAI, he co-founded Loopt, a location-based social network, and later led Y Combinator. The Dots pitch — software that keeps working after the user stops prompting it, measured by completed tasks rather than answers — extends that product sensibility into a harder problem: building something that has to be reliable enough that people trust it with ongoing work.

What We Do Not Know Yet

The launch leaves several questions open:

• Reliability. OpenAI supplies the demo scenarios. Independent evidence of how reliably Dots complete real multistep tasks — especially ones that touch multiple apps — is not yet available.
• Pricing beyond the first Dot. The first Dot is free for Pro users. OpenAI says plans will include an allowance for deeper work but did not specify the size. Multi-Dot pricing is not yet public.
• The team-of-Dots vision. OpenAI says it envisions teams of Dots working together, but the launch is limited to starting with one primary Dot.
• The iMessage and RCS rollout. Text messaging support is listed as "coming soon" for Pro users who join a waitlist.
• Safety track record on GPT-6 Astra. The model has a documented ExploitBench rating. How that translates to a persistent agent acting across a user's apps is not yet tested publicly.

Named Sources

Primary source: OpenAI's announcement post at openai.com/index/introducing-dots/ (September 29, 2026).

Named coverage: TechCrunch (Lucas Ropek, September 29, 2026), WIRED (Reece Rogers, September 29, 2026), RuntimeWire (Ryan Merket, September 29, 2026), 9to5Google, CellCog, Digital Trends, DataCamp, LinkedIn, X/Twitter (OpenAI and Sam Altman).

Named sources in coverage: Sam Altman (OpenAI CEO, DevDay keynote and X announcement).

Unverified claims: The specific reliability of Dots at completing multistep tasks is based on OpenAI's own demo scenarios, not independent testing. Multi-Dot pricing and the iMessage/RCS rollout timeline are described as future plans, not current facts. The GPT-6 Astra safety context is based on prior reporting (ExploitBench "Critical" rating, covered by AIPress Post 37 on September 25) and is relevant background, not a claim about Dots specifically.

Related AIPress coverage:

• OpenAI Scraps GPT-6.1 Astra — DevDay 2026 Safety Concerns — https://aipress.blog/post/openai-scraps-gpt-6-1-astra-over-safety-concerns-day-before-devday-2026
• GPT-6 Astra Rated "Critical" on ExploitBench — the Gap Between Score and Classification — https://aipress.blog/post/gpt-6-astra-rated-critical-on-exploitbench-the-gap-between-score-and-classification
• GPT-6 Cyber at DevDay: OpenAI's Fourth Cybersecurity Model in Twelve Months — https://aipress.blog/post/gpt-6-cyber-at-devday-openais-fourth-cybersecurity-model-in-twelve-months
• OpenAI and Anthropic Dropped New Models on the Same Day — and Both Moved Toward Cheaper, Broader Access — https://aipress.blog/post/openai-and-anthropic-dropped-new-models-on-the-same-day-and-both-moved-toward-cheaper-broader-access

Sources: OpenAI (openai.com/index/introducing-dots/, September 29, 2026); TechCrunch (Lucas Ropek, September 29, 2026); WIRED (Reece Rogers, September 29, 2026); RuntimeWire (Ryan Merket, September 29, 2026); 9to5Google; CellCog; Digital Trends; DataCamp; Sam Altman X announcement.