OpenAI launches Dots: always-on personal agents to counter Meta Muse

OpenAI launches Dots: always-on personal agents to counter Meta Muse

🔎 Three weeks after Muse, OpenAI brings out the heavy artillery

OpenAI didn't give Meta time to establish its lead. At its DevDay on September 29, 2026, the company unveiled Dots: "always-on" personal agents capable of working for you continuously, each installed on its own cloud computer with its own browser. The engine is called GPT-6 Astra, and the target has never been more explicit: Meta Muse, launched three weeks earlier.

Why the rush? Because the market has become scorching hot. CNBC describes it as a "red-hot" sector, and the numbers back up the euphoria: Meta's stock jumped 29% in September alone, driven by the excitement around Muse. Google is responding with Gemini Spark, and Apple is preparing an overhauled Siri. Remaining a spectator was no longer an option.

But this launch comes amid a tense climate. A series of incidents involving OpenAI agents—on Hugging Face, around the Australian Medicare system, all the way to the freeze of GPT-6.1 Astra—has placed the safety of autonomous agents at the center of the debate. So Dots arrives with oversized safeguards, a deliberately restricted rollout... and a price tag that will raise eyebrows: $100 per month.

The Essentials

• Dots: always-on personal agents powered by GPT-6 Astra, each hosted on its own cloud computer, with its own browser.
• More than 4,000 connected apps: the agent monitors, prepares, and executes continuously, even with ChatGPT closed.
• $100/month, initially reserved for Pro, Business Premium, and Enterprise plans — not available to the general public for now.
• Enhanced security: Guardian guardrails, auto-review of actions, mandatory approvals for sensitive tasks, restricted rollout (details at Axios).
• Open warfare backdrop: Meta Muse (+29% for Meta stock in September, per CNBC), Google Gemini Spark, and Apple's revamped Siri.
• Partnership with Microsoft Agent 365 for deploying and controlling agents in the enterprise.

Recommended Tools

Dots, what exactly is it? An agent that lives on its own computer

Dots is a fleet of "always-on" personal agents: each agent runs permanently on a dedicated cloud computer, with its own browser, and is connected to more than 4,000 applications. You don't "chat" with it — you give it a mandate, and it gets to work.

The difference with ChatGPT is structural. A chatbot waits for your prompt, responds, then disappears. A Dot, on the other hand, persists: it monitors your calendar, prepares your travel, sorts your emails, tracks your orders — and chains tasks together while you do something else. Times of India sums up the product well: an always-on AI agent powered by GPT-6 Astra, built to take on Muse.

Each agent has its own machine and its own browser. This isn't an architectural detail: it's what allows the Dot to connect to your services, browse like a human, and retain its progress from one session to the next. It's also — we'll come back to this — what makes the security question eminently political.

One cloud computer per agent: isolation as a selling point

Isolation is presented as a guarantee of security: your Dot doesn't share its machine, and its actions are logged in an environment controlled by OpenAI. It's the same reasoning as the IT departments that assign a dedicated workstation to every sensitive vendor. That's expensive in infrastructure costs — and it's precisely why the subscription costs $100.

4,000 connected applications: the real moat isn't the model

The number to remember from this announcement isn't GPT-6 Astra, it's the "4,000+". Integrations — email, calendars, banks, e-commerce, travel — are what turn an LLM into an agent that's useful in everyday life. It's a moat that Meta and Google are building too, but OpenAI arrives with one of the densest integration ecosystems on the market. My take: the model makes the demo, the integrations make the product.

Why now? Because Meta validated the market in three weeks

OpenAI is making its move now because Muse has proven itself in three weeks: adoption, media buzz, and a +29% jump in Meta's stock in September alone (CNBC, September 30, 2026). The personal agents market is no longer a hypothetical scenario — it's a revenue stream the stock market has already priced in.

Muse, launched in early September, immediately disrupted the way things are done: an agent that shops on your behalf and handles your daily life, integrated into the Meta ecosystem. Amazon's response — a blunt block within hours — paradoxically reinforced its credibility: when the world's largest e-commerce player fears your agent, it means your agent is worrying someone. We covered this standoff in detail in our article on Amazon's block against Muse.

Meta didn't stop there. With Muse Spark 1.1, the company launched its first paid model and entered the agentic coding battle (our analysis). The strategy is clear: occupy every front, from the general public to developers, before the competition can get organized.

On the other side, Google is pushing Gemini Spark, deeply integrated into its ecosystem, and Apple is preparing a redesigned Siri — agentic and on-device. OpenAI, for its part, had already laid a decisive building block with GPT-Live, its full-duplex voice that allows an agent to listen and speak at the same time (our article on GPT-Live). A personal agent you can interrupt by voice, in real time, changes the experience — it's an asset no one else has at this level.

The timing of DevDay is no accident either: it's the event where OpenAI controls the narrative, with developers as its sounding board. Announcing Dots anywhere else would have risked being drowned out by Muse news; announcing it there means reclaiming the media initiative three weeks after weathering the storm.

My take: OpenAI isn't the first to arrive, but it's the only player that combines distribution (hundreds of millions of ChatGPT users), a complete tech stack, and the enterprise channel via Microsoft. That guarantees nothing, but it positions Dots as a formidable challenger — with the advantage of being able to learn from everyone else's mistakes.

The Pricing Gamble: Who Will Really Pay $100/Month for an Agent?

This is the weakest link in the announcement. Muse is free in beta, Gemini Spark is included in a Google AI Pro subscription at around $20, Siri is free — and Dots costs $100 per month, reserved for Pro, Business Premium, and Enterprise plans.

OpenAI is upfront about its calculation: agents are aimed first at those whose time carries a market price. A consultant who saves five hours a week more than justifies $100. The general public, meanwhile, won't pay — and OpenAI knows it: that's exactly why Dots isn't open to everyone from day one.

Two points to watch. First, the model pricing precedent: OpenAI has shown it knows how to slash its prices overnight — GPT-6 Sol and Luna were cut to half price, dropped 90 minutes after a competing Claude release (our article on the price war). A consumer tier at $30-40 therefore seems inevitable in 2027. Second, free competition: as long as Muse remains free of charge, every month without a consumer offering from OpenAI is another month of users hemorrhaging to Meta.

My take: this $100 is a positioning, not a final price. It's a statement of intent — a personal agent isn't a feature, it's a product that gets monetized. It sets the market at the high end, and it leaves all the room needed to move downmarket later, once infrastructure costs come down.

Security: Oversized Guardrails — and It's the Only Defensible Decision

OpenAI has deliberately over-engineered the security of Dots: Guardian guardrails, self-review of actions, mandatory approvals for sensitive tasks, restricted launch. Given the string of incidents in recent weeks, it's the only tenable stance for a product of this kind.

Axios, which dedicated its article to this aspect, details the architecture: Guardian acts as a dedicated guardrail system, self-review has sensitive actions re-checked before execution, and mandatory approvals require human sign-off for anything involving money, sending messages, or deletions. The launch itself is intentionally limited: no mass rollout until field feedback proves solid.

The context explains this caution. Recent weeks have been brutal for OpenAI's agents' reputation: an incident on Hugging Face, another around the Australian Medicare system, and the freeze of GPT-6.1 Astra. Add to that a precedent from a competitor: an agent caused Schwab to drop 67% intraday and weighed down the S&P Financials by 2% — an episode we link to Muse in our account of the flash crash. When an agent can move markets, security stops being a marketing talking point and becomes a condition of existence.

My take: for agents that pay, buy, and browse on your behalf, guardrails must be structural, not cosmetic. Mandatory approvals are the right default setting. The real question now is this: will these guardrails survive competitive pressure? On the day Muse ships more "fluid" agents with less friction, will OpenAI hold the line? That's when we'll find out whether security was a matter of conviction or just PR.

Microsoft Agent 365: consumer buzz, enterprise revenue

The partnership with Microsoft Agent 365 reveals OpenAI's real strategy: making Dots the standard enterprise agent, where $100/month per employee is a painless price point.

Agent 365 works as a control plane for agents: identity, permissions, audit logs, compliance. Concretely, an IT department will be able to provision Dots the way it provisions mailboxes: one agent per employee, defined scopes of action, traces reviewable at any time. This is exactly what it takes for an always-on agent to get past the barriers of a legal department.

The business math is crystal clear. At $100/month per seat, a company with 5,000 employees represents $6 million in potential annual revenue — from a single customer. Consumers drive the buzz and adoption; enterprises drive recurring revenue. It's the same playbook that made Office a success, and then Microsoft 365.

One gray area remains: Microsoft is pushing its own Copilot agents in the same console. OpenAI brings the models, Microsoft the distribution — but in the long run, the two players will fight over the orchestration layer. Companies would do well to negotiate the portability of logs and configurations in the very first contract, not at the moment of exit.

Should You Take the Plunge? My Take After the DevDay

If you're already a Pro or Business subscriber and your weeks look like a pile-up of repetitive tasks, give Dots a try — with validations enabled and a deliberately narrow scope. Otherwise, wait for the public rollout: there's no rush, and the product will evolve fast.

Start with low-risk tasks: competitive monitoring, email triage, meeting prep, case follow-up. Hold off on payments and shipments until the validation friction level is properly calibrated. A poorly scoped agent always costs more than a subscription.

If the $100 price tag puts you off, self-hosting is a credible alternative. Open source agents running locally with Ollama already cover part of these needs (our guide), and self-hosted models like Kimi K2.6 from Moonshot AI or GLM-5 from Z.AI are sufficient for common agentic tasks — our comparison of LLMs for agents details these choices. A VPS from Hostinger starts at around $5/month and runs 24/7. You'll lose OpenAI's 4,000 integrations; you'll gain control over your data.

For my part, I see Dots as the most complete agent product on the market — and as a real-world test. If users pay up, the agent war enters its economic phase. If nobody pays, Muse will have won the narrative and OpenAI will have to slash its prices. Either way, the power is in your hands: judge the agent by what it delivers, not by what it promises.

❌ Common Mistakes

These four mistakes come up with every agent launch — avoid them from day one.

Mistake 1: confusing an always-on agent with an enhanced chatbot

A Dot doesn't wait for your prompt: it plans, follows up, and acts. Many will treat it like ChatGPT and be surprised by actions taken without them. The solution: check the activity log every day during the first week, and set up checkpoints on anything that goes beyond your scope.

Mistake 2: handing over all the keys from day one

Payments, work emails, social media: mandatory validations exist for a reason. The solution: start with a narrow scope (monitoring, triage, preparation) for two weeks, then expand gradually, one sensitive action at a time.

Mistake 3: believing the advertised safeguards are enough

Guardian and auto-review are marketing promises, and recent incidents — Hugging Face, Australian Medicare — show that agents go off the rails even at the best of them. The solution: keep a human in the loop for any irreversible action and revoke access that is no longer needed.

Mistake 4: subscribing at $100/month without a quantified use case

If you can't list three repetitive tasks the agent will take over this week, the subscription isn't justified. The solution: map out your needs first using the free offerings (Muse, Gemini Spark), then measure the time actually saved before pulling out your credit card.

❓ Frequently Asked Questions

The questions that keep coming up after the announcement.

When will Dots be available to the general public?

OpenAI hasn't announced a date. The launch is restricted to Pro, Business Premium and Enterprise plans ($100/month), with a gradual rollout according to Axios. Expect several months before a wider opening, and keep an eye on OpenAI's official page for updates to the timeline.

What's the difference between Dots and Meta Muse?

Dots targets professionals: an always-on agent on a dedicated cloud computer, over 4,000 apps, $100/month, Guardian safeguards. Muse targets the general public with a free agent integrated into the Meta ecosystem, focused on everyday life and shopping — a positioning that has already earned it a deliberate block from Amazon. Two products, two markets, two business models.

Can Dots really act without me at all times?

Yes, that's the "always-on" principle: the agent runs on its own cloud computer and chains tasks continuously. But sensitive actions — payments, shipments, deletions — require human validation according to OpenAI. In practice, you remain the supervisor: the agent handles the routine, you approve the risky points.

Can you have an always-on agent without paying $100/month?

Yes, by self-hosting an open source agent with Ollama on a VPS — our guide details the steps to follow. A server at Hostinger starts around $5/month and runs 24/7. You'll lose OpenAI's 4,000 integrations and some of the convenience, but you keep full control over your data and your costs.

Are personal agents safe in 2026?

Better than six months ago, but not safe. Recent incidents — Hugging Face, Australian Medicare, the Schwab flash crash linked to an agent — have proven it. Guardian-style safeguards and mandatory validations reduce the risk, they don't eliminate it: never grant financial access without human supervision, even to a "certified" agent.

✅ Conclusion

With Dots, OpenAI is making official what Muse had started: 2026 is the year personal agents become a paid product — and at $100/month, the bet is that your time is worth more than your subscription. Before picking your side, compare the players in our ranking of the best autonomous AI agents.