The last two weeks have been great for the AI world. Meta's Muse and OpenAI's Dots promise to keep working after you close the tab. That has been the focus of the last few weeks of AI product news, rather than better answers in a chat box. I already run a version of this on Grok Bot, and my experience with it shapes how I see these launches.

I still open ChatGPT, Claude, or Perplexity when I want an answer, a draft, or a search with sources. Perplexity finds things and shows the source. Claude is who I trust when a document needs a careful pass and my general workbench. A long thread and a remembered preference are useful, but these remain session tools. When I leave, the work waits until I come back and ask again.

An always-on agent has its own computer and can use the apps I connect. It is supposed to come back with progress or a question, without waiting for me to reopen the conversation. Muse and Dots are selling that arrangement as a personal workforce.

Muse launched on 8 September 2026. Meta describes it as an agent that does the work instead of stopping at an answer. It uses Muse Spark and runs on Muse Secure VM, a dedicated cloud computer with its own browser. You message it through the Muse app or WhatsApp, and longer tasks continue after you close the app. Before sending an email or making a payment, it is supposed to check with you. Payments can go through Link, Stripe's wallet, using a one-time card so your real card number stays off the merchant page.

You choose which apps Muse can access, and email access can be read-only. The rollout is in the US, on iPhone, Android, and muse.ai. It is a consumer agent for errands, life admin, and shopping, with one broad brief: you state a goal once, and it keeps working on it. But it cannot necessarily shop wherever you want. Amazon has blocked Muse, arguing that Meta did not get permission and that the agent does not identify itself. That lockout affects what Muse can actually do.

Dots launched at OpenAI's DevDay on 29 September 2026. A Dot is an always-on agent using GPT-6 Astra, with its own cloud computer and browser and plugins for more than 4,000 apps. You can reach it in ChatGPT, Slack, or Teams, and by voice. You start with one primary Dot. Teams of Dots and specialist Dots for companies are planned for later. The rollout covers Pro, Business Premium, and Enterprise in eligible markets.

Your laptop stays off unless you connect it. Idle "proactive research" is read-only; assigned work can go further under rules that allow an action, require approval, or block it. You can also open the Dot's computer and look at what it is doing. Dots is aimed at ongoing professional work around ChatGPT, with Slack and Teams as other ways to reach the same agent. At launch, that means one agent with a broad work brief, rather than a team of specialists.

On Grok Bot, I have been building a staff with separate responsibilities. I have a CFO, with two agents under it looking after my finances. A CMO helps with email and social posts. A personal trainer is next, although health is still a plan. I am leaving the setup details out of this note because the division of work is what interests me.

Muse, Dots, and my staff are all meant to keep a job going over time, on computers of their own. In their descriptions, they also put approval before costly mistakes such as sending, paying, or making changes you cannot undo. Each could become more useful as it learns how I like things done. That learning also creates risk.

The appeal of Muse and Dots is that they are products you can open. The computer, permission model, and rollout come with the package; you do not have to invent an org chart on day one. Muse focuses on life admin and checkout. Dots offers ongoing project work beside ChatGPT, with a computer you can inspect. Both are easier to switch on than a staff I have to design myself.

I prefer the extra setup because I can keep permissions tied to a job. Finance stays with the CFO and the two agents under it. Email and social drafts stay with the CMO. The agent drafting a post does not need finance logins, and the agent preparing a finance note does not need to send mail as me. "Ask the CFO" is a clearer instruction than "you know my life, sort it out." Money, publishing, and health do not have to share one memory or one set of credentials. Keeping them separate makes the setup easier for me to trust.

That does not remove the risks. A shopping agent can buy the wrong item or try to shop at a store that has refused it. An approval card does little if I tap through it like a cookie banner. Amazon's block on Muse is a practical limit: I can send the agent, but the merchant can still say no.

A work agent connected to thousands of apps can make small messes for a month. Keeping idle research read-only is the right default. I worry more about assigned work that nobody thought to write a rule for. My staff has its own version of this problem: I can become careless because an agent has a narrow job title. Email may be a narrow subject, but permission to send as me is still broad. Calling an agent a CMO does not limit what it can do; its permissions have to do that.

There is also the question of who holds the memory and credentials. Muse keeps them in Meta's VM. A Dot's computer sits with OpenAI. My staff sits with the assistant I chose. I can tell an agent to forget something, but I still have to remember to say it.

I am staying with the staff. My week already splits into money, what I publish, and health, which I am only now making room for. One agent handling all of it would be faster to set up and harder for me to trust. Muse and Dots have brought this kind of always-on workforce out of the lab and into public products. They offer a broad agent now; I prefer several agents whose access stays limited to their jobs.

If you have been watching these launches from the outside, which arrangement looks more usable to you, and why?

Sources

Meta, Introducing Muse (8 September 2026): about.fb.com/news/introducing-muse-personal-ai-agent

OpenAI, Introducing dots (29 September 2026): openai.com/index/introducing-dots

The Verge, Amazon blocks Meta's Muse: www.theverge.com/tech/amazon-blocks-meta-muse-ai-agent-shopping

"Gork Bot, Muse & Dots": A new battle with the "Always-On AI workforce"

Job breakdowns

Job breakdowns

Command Menu