OpenAI unveiled a new class of AI agent called Dots on September 29, 2026, and the framing was unmistakable: this is OpenAI’s answer to Meta’s fast-growing Muse agent, and the opening move in a fight over who controls the enterprise AI agent market. Reuters described the launch as “OpenAI takes on Meta with always-on Dots agent in enterprise AI push,” while Axios called the product OpenAI’s answer to Meta’s viral Muse. The New York Times ran a similar framing, positioning Dots as a direct rival not just to Muse but to Google’s Gemini Spark and an AI-agent startup called Instinct.
The core idea behind a Dot is simple to describe and much harder to build: instead of answering one prompt at a time, a Dot works toward a goal continuously, in the background, checking in with a human only when it needs a decision or approval. OpenAI says each Dot runs on its own cloud computer, connects to more than 4,000 third-party apps, and can be reached through ChatGPT, Slack, or Microsoft Teams. It is powered by GPT-6 Astra, the model OpenAI has been positioning as its strongest system for computer-use tasks — the kind of work that involves clicking through software the way a person would rather than just generating text.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What OpenAI’s Dots Agent Actually Does
According to OpenAI’s own announcement and reporting from Axios, a Dot starts with a goal a user assigns it and a set of boundaries defining what it is allowed to do without checking back in. From there, the agent keeps working: researching, drafting, updating documents, and building software as the underlying project changes. Reuters gave a concrete example that OpenAI used to illustrate the concept — a Dot updating a sales proposal automatically when a customer’s requirements shift, and then building a working demo for the sales team to review before it goes out.
That is a meaningfully different pitch than the assistant model most people associate with ChatGPT. A standard chatbot session ends when the user closes the tab. A Dot, according to OpenAI, keeps running, learns from feedback over time, and treats a task as an ongoing responsibility rather than a single exchange. CNBC’s live coverage of the DevDay unveiling noted that Dots can already reach more than 4,000 connected apps, with text-message access described as “coming soon” on top of the Slack, Teams, and ChatGPT channels available at launch.
Built on GPT-6 Astra
The model doing the heavy lifting inside every Dot is GPT-6 Astra, which OpenAI released in September 2026 and which multiple outlets — Reuters, the New York Times, Axios, Business Insider, and CNBC among them — identified as the engine behind the new agents. Notably, the New York Times reported that OpenAI withheld a planned successor, GPT-6.1 Astra, from that same September 2026 rollout over internal safety concerns, leaving the original Astra model as the one actually powering Dots at launch. Business Insider described Astra as OpenAI’s strongest model yet for computer use, capable of moving through apps and websites in a way that resembles how a person would navigate them, rather than relying purely on API calls. That distinction matters for enterprise buyers, since a large share of business software still has no clean API and can only be operated the way a human employee would operate it — by clicking, typing, and reading a screen.
OpenAI has also been widening its model lineup on the pricing side. As we covered when GPT-6 Sol and Luna undercut Claude Opus 5.5 on price, the company has been using cheaper Astra-family variants to compete on cost while reserving its most capable systems for premium, high-stakes work like the always-on agent behind Dots.
Codex and ChatGPT Work Under the Hood
Reuters and Axios both reported that Dots draw on OpenAI’s Codex coding tool and its ChatGPT Work product to research topics, analyze data, prepare documents, and write software as part of completing a goal. That combination — a persistent agent, a coding tool, and a workplace-document tool, all running on one model — is OpenAI’s attempt to collapse several separate product categories into a single subscription. It also extends a pattern the company has been building toward for months. When ChatGPT Voice added access to email, calendar, and Slack earlier this year, it was a preview of the same philosophy: fewer standalone tools, more integration into a single agent that reaches across a company’s existing software stack.
Who Gets Access First
OpenAI’s rollout plan, as described in its announcement and confirmed across Axios, Business Insider, and CNBC’s coverage, starts narrow. Access is going first to ChatGPT Pro subscribers and Business Premium and Enterprise customers in eligible markets, with the New York Times noting that the EEA, Switzerland, and UK were specifically excluded from the initial September 2026 release. Each user gets one Dot to start, though Axios reported that OpenAI’s stated long-term vision is teams of Dots working together on shared projects rather than a single agent per person.
As of OpenAI’s October 2026 Help Center documentation, a user’s first Dot costs $0 extra on top of an existing ChatGPT Pro, Business Premium, or Enterprise subscription, confirming that the agent is bundled into those tiers rather than sold as a separate line item. That is a notable contrast with how some rivals have approached agent pricing, and it suggests OpenAI is treating Dots as a retention and upsell feature for its highest-paying subscribers rather than a new standalone product line, at least for now.
The Meta Muse Rivalry Behind the Timing
The timing of the Dots launch is not a coincidence. Meta’s own always-on-style agent, Muse, generated what Reuters described as millions of downloads earlier in September, a surge that rattled shares of companies investors view as either beneficiaries or potential casualties of the autonomous-agent trend. We tested that momentum directly in our own comparison piece, Meta Muse Tops ChatGPT-6 3-2 in a 5-Prompt Test, which found Muse outperforming OpenAI’s previous-generation model on a majority of head-to-head prompts — a result that likely added urgency inside OpenAI to answer with something more capable than an incremental ChatGPT update.
Muse has not been without its own problems. Security researchers flagged issues covered in our report on the Meta Muse zero-day that exposed more than 50 commands through a single setting, and the flaw reportedly went unpatched for at least 24 hours after disclosure. That history gives OpenAI an opening to position Dots as the more carefully governed alternative — assuming its own boundary-and-approval system holds up under real-world enterprise use, which has not yet been independently tested at scale.
Meta Isn’t Standing Still Either
One day before OpenAI’s announcement, Meta detailed its own enterprise ambitions. According to reporting reviewed alongside the Dots launch, Meta unveiled a Meta Enterprise Platform on September 28, 2026, intended to bring its AI models and agents — including Muse and a separate Meta Business Agent — directly to corporate customers. Meta also named Chirantan “CJ” Desai as chief enterprise platform officer, reporting to Mark Zuckerberg, a signal that the company is building a dedicated enterprise sales organization rather than treating business customers as a side effect of its consumer AI push. We broke down the announcement in more detail in Meta Enterprise Platform: Zuckerberg’s $145B AI Bet and followed up with What Meta Enterprise Platform Actually Includes.
That spending context matters. Zuckerberg has already committed the company to an aggressive AI infrastructure build-out, a bet we detailed in our coverage of Zuckerberg’s $145 billion AI spending plan with no clearly stated fallback. Meta’s advantage in this fight is distribution — billions of existing users across Facebook, Instagram, and WhatsApp, plus the viral momentum Muse has already built. OpenAI’s advantage is the opposite: a business software ecosystem already wired into Slack, Teams, Codex, and thousands of third-party apps that companies use every day. Whichever company can convert its existing strength into enterprise contracts fastest is likely to set the pace for this entire product category over the next year.
How Dots Compares to Muse, Gemini Spark, and Instinct
The New York Times named three specific rivals Dots is meant to compete against: Meta’s Muse, Google’s Gemini Spark, and an AI-agent startup called Instinct. Here is how the publicly reported details stack up so far, based on OpenAI’s own announcement and the outlets that covered it.
| Agent | Maker | Underlying Model | Access Channels | Initial Tier Access |
|---|---|---|---|---|
| Dots | OpenAI | GPT-6 Astra | ChatGPT, Slack, Teams (text coming soon) | Pro, Business Premium, Enterprise |
| Muse | Meta | Meta’s in-house model family | Meta apps, standalone app | Consumer-first, enterprise via Meta Enterprise Platform |
| Gemini Spark | Gemini model family | Google Workspace ecosystem | Not detailed in current reporting | |
| Instinct | Independent startup | Not disclosed in current reporting | Not detailed in current reporting | Not detailed in current reporting |
The gaps in that table are deliberate — reporting on Gemini Spark and Instinct is thin so far compared to the detail available on Dots and Muse, and we are not filling in numbers that have not actually been confirmed. What is clear is that OpenAI, Meta, and Google are now all racing toward the same basic product shape: an agent that persists across sessions, has some autonomy within defined limits, and plugs into the software a business already runs on.
From ChatGPT Enterprise to Dots: OpenAI’s Product Path
Dots did not appear out of nowhere. It is the latest step in a progression OpenAI has been building for roughly three years, moving steadily from a conversational chatbot toward something closer to a digital employee. ChatGPT Enterprise established the baseline: business-grade access, admin controls, and security features aimed at IT departments rather than individual users. GPTs came next, letting organizations configure custom versions of ChatGPT tuned to specific workflows. Then came Agents — systems that could use tools, browse the web, and operate a computer rather than just produce text. Dots is the point where those threads converge into one persistent, goal-driven system.
| Product | Core Shift | Interaction Model |
|---|---|---|
| ChatGPT Enterprise | Business-grade access and admin controls | Prompt-and-response, session based |
| GPTs | Custom configurations per workflow | Prompt-and-response, tailored persona |
| Agents / Computer Use | Tool use and operating software directly | Task-based, still largely single-session |
| Dots | Continuous, goal-driven operation | Always-on, checks in only when needed |
OpenAI has also been building out a security-focused model track alongside its consumer and enterprise agents, a move we covered when the company was preparing GPT-6 Cyber as its fourth dedicated security model. That parallel investment suggests OpenAI is at least aware that an agent with standing access to thousands of connected apps creates a larger attack surface than a chatbot that only responds when prompted — a concern that will likely shape how quickly cautious enterprise IT departments actually adopt Dots.
Market Reaction and What Investors Are Watching
The financial picture around Dots is still forming. Reuters reported that Muse’s download surge earlier in September had already jolted shares of companies investors believe stand to gain or lose from the rise of autonomous AI agents, and separate reporting indicated Meta shares came under pressure amid concern that a rumored OpenAI agent could threaten Meta’s own AI-driven rally. Neither Reuters nor the other outlets covering the Dots launch itself have published a confirmed post-announcement stock move, an enterprise customer count, or an OpenAI valuation tied specifically to this product, so treat any number beyond the reported app-connection count and rollout tiers as unconfirmed for now.
What is measurable is the pattern of competitive spending. Meta’s enterprise ambitions sit inside a much larger capital commitment, and OpenAI’s own infrastructure needs to support millions of always-on agents running in parallel are not trivial either — each Dot, by OpenAI’s own description, gets its own cloud computer rather than sharing a lightweight session. That is a meaningfully different cost structure than serving a chatbot response, and it is one reason analysts will be watching how OpenAI prices Dots once the current bundled-access period ends.
Security and Oversight Questions Enterprises Will Ask
Handing an AI agent standing access to Slack, Teams, and thousands of connected apps raises the obvious question of what happens when it makes a mistake, or when its access is misused. OpenAI’s pitch is that Dots operate within boundaries a user defines up front and check in when a decision falls outside those limits — a permission model rather than a fully autonomous one. A simplified version of how that boundary configuration might look in a business setting is illustrated below, based on the general access-and-approval model OpenAI has described publicly.
{
"dot_goal": "Keep the Q4 sales proposal in sync with customer requirements",
"connected_apps": ["Slack", "Microsoft Teams", "Salesforce"],
"autonomy_boundaries": {
"can_edit_documents": true,
"can_send_external_email": false,
"requires_approval_for": ["contract terms", "pricing changes"]
},
"check_in_channel": "Slack"
}That kind of configurable boundary is exactly the feature that will be tested hardest once Dots reach real enterprise environments. Meta’s own experience is a cautionary example: the Muse backdoor issue we reported on remained unpatched for at least a day after it became public, which is a long window for an agent with access to sensitive workplace systems. IT and security teams evaluating Dots will likely want clear answers on patch response times, audit logging, and whether a Dot’s access can be revoked instantly if something goes wrong — none of which has been detailed in the launch coverage so far.
Historical Context: Why Enterprises Are the New Battleground
Consumer chatbots were the first phase of the generative AI boom, and that market has largely matured into a handful of dominant players trading market share back and forth on benchmark scores. Enterprise software is a different game entirely — longer sales cycles, higher contract values, and stickier customer relationships once a company has wired an agent into its internal tools. That is exactly why Meta, a company built almost entirely on consumer advertising, is now hiring a chief enterprise platform officer and building a dedicated business arm around Muse and Meta Business Agent. It is also why OpenAI, which built its early lead on consumer ChatGPT adoption, is pushing Dots toward Business Premium and Enterprise tiers first rather than rolling it out broadly to free users.
Both companies are chasing the same prize: becoming the default AI layer that a company’s other software has to plug into. Whoever wins that position captures not just subscription revenue but a level of platform lock-in that is much harder to dislodge than a chatbot habit. That is the real stake behind what looks, on the surface, like a product feature race.
What Developers and IT Teams Should Do Now
For engineering and IT leaders evaluating whether to pilot Dots, the practical first step is scoping access narrowly. Given that each Dot gets its own cloud computer and standing access to connected apps, the safest early rollout is a single low-stakes workflow — internal documentation updates or research tasks — rather than anything touching customer data, contracts, or financial systems until the security model has been tested in production. Enterprise admins should also confirm whether their existing ChatGPT Enterprise or Business Premium contract already includes Dots access or requires a separate upgrade, since the rollout tiers reported so far are Pro, Business Premium, and Enterprise rather than every paid plan.
Teams already running Codex-based workflows have a natural entry point, since Dots reportedly draws on the same coding tool for software-related tasks. That overlap could make Dots an easier sell to engineering organizations that already trust OpenAI’s coding stack than to departments with no existing OpenAI footprint at all.
Predictions: Where the Agent War Goes Next
Based on the pattern established by this launch and Meta’s parallel enterprise push, a few directions look likely over the next six to twelve months.
- OpenAI will almost certainly expand Dots beyond “one per user” toward the multi-agent teams Axios says the company has already described as its long-term vision, likely tied to higher-priced Enterprise tiers.
- Meta will move quickly to formalize enterprise pricing and support for Muse and Meta Business Agent under its new Meta Enterprise Platform organization, using Chirantan Desai’s hire as the anchor for a dedicated sales motion.
- Security incidents tied to always-on agents — following the pattern of the Muse backdoor disclosure — are likely to become a recurring story in 2026 and 2027 as more companies grant these systems standing access to internal tools.
- Google will need to clarify what Gemini Spark actually offers relative to Dots and Muse, since current reporting leaves its enterprise agent positioning notably vague compared to its two biggest rivals.
- Pricing for standalone agent access, separate from bundled ChatGPT and Meta subscriptions, is likely to emerge within a year as usage data reveals how expensive it actually is to run millions of persistent, cloud-computer-backed agents at scale.
The Bigger Picture
What makes the Dots launch notable isn’t a single spec or feature — it’s the signal that two of the largest AI companies in the world now agree on what the next product category looks like. Both OpenAI and Meta are betting that businesses want an agent that keeps working after the prompt window closes, not just a smarter version of the chatbot they already have. Coverage from CNBC, Business Insider, Wired, The Verge, and TechCrunch all converge on the same read: this is the opening round of a much longer contest, and the rollout tiers, security track record, and pricing decisions made over the next few months will matter more than the DevDay keynote itself.
Frequently Asked Questions
What is OpenAI’s Dots agent?
Dots is an always-on AI agent OpenAI announced on September 29, 2026, designed to work toward a user-defined goal continuously rather than responding to single prompts. It runs on GPT-6 Astra, uses its own cloud computer per user, and connects to more than 4,000 third-party apps through ChatGPT, Slack, and Microsoft Teams.
How is Dots different from a regular ChatGPT agent?
Standard ChatGPT agents typically complete a defined task within a session. A Dot is designed to persist across sessions, keep working in the background, adapt to changing project requirements, and reach out to the user only when it needs approval or clarification, according to OpenAI’s announcement.
Who can access OpenAI Dots right now?
Access is rolling out first to ChatGPT Pro subscribers and Business Premium and Enterprise customers in eligible markets, with the EEA, Switzerland, and UK initially excluded, according to the New York Times. Each user starts with one Dot, and as of OpenAI’s October 2026 Help Center, that first Dot costs $0 extra on top of an existing subscription.
How does Dots compare to Meta’s Muse?
Muse is Meta’s own always-on-style agent, which gained millions of downloads earlier in September 2026 according to Reuters. Dots is OpenAI’s direct response, aimed at enterprise customers first rather than consumers, with deeper integration into workplace tools like Slack and Teams from the outset.
What model powers Dots?
Dots runs on GPT-6 Astra, which OpenAI and outlets including Business Insider describe as the company’s strongest model for computer-use tasks, meaning it can navigate apps and websites in a way that resembles human interaction rather than relying only on structured API calls.
Does Dots cost extra on top of a ChatGPT subscription?
Current reporting does not include a separate standalone price for Dots. It is being bundled into existing ChatGPT Pro, Business Premium, and Enterprise plans rather than sold as an independent product at launch.
Is Dots secure enough for sensitive enterprise data?
OpenAI has described a boundary-and-approval system that limits what a Dot can do without human sign-off, but the model has not yet been tested at scale in production enterprise environments. Rival agent Muse faced a disclosed security flaw that reportedly went unpatched for at least 24 hours, underscoring why security teams are likely to move cautiously with any always-on agent.
What other companies are building competing agents?
Beyond Meta’s Muse, the New York Times named Google’s Gemini Spark and an AI-agent startup called Instinct as competitors in the same category, though public details on both remain limited compared to the specifics available for Dots and Muse.
Related Coverage
- Meta Muse Tops ChatGPT-6 3-2 in 5-Prompt Test [2026]
- Meta Beats OpenAI to AI Devices by 1 Week [2026]
- Meta Muse Adds Avatars, Mac Control in 15 Days [2026]
- Meta Enterprise Platform: Zuckerberg's $145B AI Bet [2026]
- Muse Spark 1.3 Enters Frontier Tier, 3.4M Downloads [2026]
