Dots, Cue, Muse: the always-on AI agent era for SMEs

In three weeks, three AI giants launched the same idea: always-on AI agents that keep working even after you close the app. On September 8, 2026, Meta unveiled Muse. On September 28, the startup Manus launched Cue. On September 29, OpenAI presented dots at its DevDay in San Francisco. For an SME leader, this convergence is not a technical detail: it marks the shift from AI that answers to AI that acts on its own, continuously, with standing access to your tools. Here is what these always-on agents actually enable, what they cost, and the precautions to take before connecting one to your business.

In brief

• OpenAI launched dots on September 29, 2026: agents that run on their own cloud computer, connected to more than 4,000 apps, reachable from ChatGPT, Slack or Teams.
• Manus launched Cue on September 28, 2026: each agent gets its own phone number, its own email address, and a wallet to pay on your behalf.
• Meta launched Muse on September 8, 2026: a personal agent that books, fills out forms and completes purchases, currently available in the United States.
• All three products share the same principle: you set a goal once, and the agent keeps executing it without a new prompt, including while you are offline.
• For an SME, the question is no longer just "which tool to pick" but "how much standing access am I willing to grant", since these agents keep a memory and an access that survive a simple disconnection.

Three launches in three weeks

This tight timeline shows a race between the major AI vendors on a new front: the agent that never stops.

This is not a scheduling coincidence: all three vendors are responding to the same observation, confirmed by OpenAI at its September 29 DevDay, namely that enterprise AI usage is shifting from one-off chat to routines handed off once and for all.

What is an always-on AI agent?

An always-on AI agent differs from a classic work assistant in three ways. First, it has its own execution environment (a "cloud computer" at OpenAI and Meta), separate from your device: it keeps running even with your computer off. Second, it retains a memory of past exchanges and decisions, which it reuses without being asked again. Third, it holds standing access to a set of apps and channels (messaging, calendar, and for Cue, its own phone number), letting it act without waiting for a new instruction.

Concretely, at OpenAI, a dot runs on the GPT-6 Astra model (launched September 3, 2026) and is controlled from ChatGPT, Slack, Teams or by voice. You give it a standing goal, for instance watching customer feedback and proposing small fixes, and it keeps running that loop without a new prompt. Each action follows a rule the user sets: act on its own, act on its own if pre-approved, ask first, or hand off the decision to a human.

Dots, Cue, Muse: the comparison

All three products target the same use case, but with different philosophies on autonomy and the access granted to the agent.

That last row deserves a leader's attention: Meta chose to add a separate oversight agent that reviews every action before it goes out on the web, a safeguard the other two vendors have not announced in this form to date.

What it actually costs

At OpenAI, the first dot is included with the Pro 100, Pro 200 or Pro 500 plans ($100, $200 or $500 a month) or with a Business Premium seat. OpenAI has not disclosed pricing beyond the first month of usage: treat the included dot as a trial period, and measure it before adding a second one. At Manus, the base subscription starts at $20 a month, with tiers at $40 and $200, plus $30 to $50 for each additional "cloud computer", billed separately from the subscription. Cue itself remains free in early access, with no announced pricing beyond that. Muse is free for most uses, with paid plans for doing more.

Risks to know before connecting an always-on agent

Three points deserve attention before granting standing access to one of these agents.

The first concerns access that outlasts the moment. OpenAI says so explicitly in its documentation on dots: a dot runs proactive research across everything connected to it and forms memories from what it reads, even when never asked to. Disconnecting an app later does not erase what it already learned: only deleting the agent entirely does.

The second concerns an own identity that receives messages. With Cue, an agent has its own phone number and mailbox: anyone can therefore reach it directly. Before switching on a routine, decide what an incoming message is allowed to trigger for that agent.

The third concerns separate billing. Manus's "cloud computers" are billed on top of the subscription, and their files are deleted if the subscription lapses. It is worth building an exit plan before depending on an agent's phone number or files.

How to adopt an always-on agent safely in your SME

FAQ

What is an always-on AI agent compared to a classic work agent?

A classic work agent (like ChatGPT Work or Claude Cowork) runs a one-off mission and then stops. An always-on agent like dots, Cue or Muse keeps an active goal continuously, with open access to your apps, until you explicitly remove it.

Are always-on agents already available in Europe?

OpenAI's dots is available for Business Premium subscribers in all supported ChatGPT regions; the Pro plan, however, excludes the European Economic Area, Switzerland and the United Kingdom at launch. Meta's Muse is for now limited to the United States. Manus's Cue works by invitation, with no geographic restriction announced at this stage.

How much does an always-on AI agent cost for an SME?

Prices range from free (Muse, Cue in early access) to several hundred dollars a month for OpenAI's most complete plans. Watch for extra costs: Manus's "cloud computers" are billed separately from the subscription, and OpenAI has not yet disclosed pricing for usage beyond the first month.

What is the main risk for an SME connecting an always-on agent?

The main risk is access that lasts longer than expected: the agent's memory and its access to connected apps persist even after a disconnection, as long as the agent itself is not deleted. Hence the value of starting with a low-stakes routine and checking the provider's data policy before any sensitive connection.

Conclusion

Three launches in three weeks, at OpenAI, Meta and Manus, confirm the same direction: AI is moving from one-off chat to the always-on agent, which keeps an active goal and open access continuously. For an SME, the opportunity is real (automatic monitoring, message sorting, recurring tasks delegated for good), provided you move with method: a low-stakes routine, a cautious action rule, and an exit planned before you enter. To go further on adopting AI agents in business, explore our resources and success stories.