Always-On AI Agents: What Dots, Muse and Grok Bot Mean for Business

In the space of seven weeks, three of the biggest AI companies have started selling something new: AI workers that keep going after you log off. xAI launched Grok Bot in August, Meta launched Muse in September, and today OpenAI launched Dots at its DevDay event. Meta also released a version of Muse built for small businesses on the same day.

These agents don’t wait for your next question. You give them a goal, they sign in to your tools and work on it around the clock, and they come back when they need a decision. For a small team that’s short on hours, the appeal is obvious. So are the questions: what should one of these actually be allowed to do, and what happens when your customers start sending their own agents to your website?

What is an always-on AI agent?

An always-on AI agent is AI software that works toward a goal in the background rather than answering one prompt at a time. The three new products share a few traits:

• Their own computer in the cloud. Each agent has a virtual machine (a computer running in a data center) with a web browser, so it can keep working when your laptop is closed.
• Access to your tools. You connect the apps it can use, such as your email, calendar, store, CRM or accounting software, and it signs in the way an employee would.
• Memory and feedback. They learn how you like things done and carry that context from one task to the next.
• Check-ins instead of step-by-step prompts. They work on their own and return when something needs your approval.

If you read our earlier guide to AI agents for small business, the difference is time. Those agents run a task when asked. These ones keep running.

Grok Bot, Muse and Dots compared

Sources: xAI’s Grok Bot announcement, Meta’s Muse launch, Meta’s Muse for Small Business announcement, and coverage of Dots from 9to5Google and SiliconANGLE.

Demand is real. Muse reached number one on the US App Store 10 days after launch, and Sensor Tower estimated more than 3.4 million downloads by September 25. Dots is the most restricted of the three for now: SiliconANGLE reports that Pro subscribers in the European Economic Area, Switzerland and the UK are excluded at launch, and pricing for extra Dots hasn’t been announced.

What can an AI worker do for a small business?

The launch examples are a useful guide because they are specific. Meta says Muse for Small Business can analyze sales and campaigns, prioritize email and calendar, draft ad campaigns and flag suspicious expenses. xAI lists outbound sales research, CRM upkeep, customer follow-up and invoice processing. SiliconANGLE’s report on Dots mentions invoice preparation and investigating bugs raised in Slack.

In practice, the jobs that suit an always-on agent share two things: they happen continuously, and nobody on your team enjoys doing them. Good candidates include:

• Watching your inbox and store for orders, refunds or complaints that need a human, and drafting the reply.
• Keeping your CRM tidy after every call and email.
• Chasing overdue invoices with polite, escalating reminders you approve.
• Pulling a weekly summary of sales, ad spend and site traffic into one message.
• Monitoring competitors’ prices or new product launches.

Jobs that involve judgment, money or promises to customers are different. The agent can prepare the work. A person should make the call.

The risks the launch videos skip

An agent that can sign in to your accounts can also make mistakes in them, quickly and at 3 a.m. A few risks deserve attention before you connect anything:

• Too much access. The easiest setup is to hand over your own logins. That gives the agent everything you can reach, including payroll, customer data and admin settings.
• Prompt injection. This is when instructions hidden in an email, web page or document trick an agent into doing something you didn’t ask for, such as forwarding files. Agents that read your inbox all day are exposed to it constantly.
• Accountability. If the agent emails a customer the wrong price, that’s still your business’s promise.
• Unknown costs. Launch pricing is generous and incomplete. Budget for the plan you’ll need once usage limits bite.

The companies themselves are candid about this. NBC News reported that OpenAI launched Dots a day after apologizing for an incident in which an internal model under development accessed non-public information related to Australia’s public health insurance program. OpenAI says that model was not on track to ship, but it’s a reminder that capable agents go where their access lets them.

How to trial an AI worker safely

A sensible first month looks like this:

1. Pick one goal you can measure. “Follow up every unpaid invoice older than 14 days” is better than “help with admin”. Record how many hours it takes today.
2. Create a separate account for the agent. Give it its own email address and user seat in each tool, with the lowest permission level that still gets the job done.
3. Write the approval rules down. Anything that sends, publishes, spends or deletes comes back to a person, at least until you trust its work. All three products support this.
4. Keep it away from what can’t be undone. Banking, payroll, contracts and admin settings stay off limits during the trial.
5. Review its activity every week. Read what it did, not only what it reports. Note every correction you make.
6. Measure again after 30 days. Compare hours saved and errors against your baseline, then decide whether to expand, adjust or stop.

It sounds like a lot of process for a free trial. It’s the same process you’d use for a new hire with keys to your systems, and it’s what separates useful agents from the projects that quietly get switched off.

Your customers are hiring AI workers too

There’s a second side to this that most coverage misses. Muse and Dots aren’t only for running your business. People use them to shop, book and compare for them, which means agents will increasingly arrive at your website instead of people.

That shift is already causing friction. Around September 21, Amazon blocked Muse from shopping on its store, saying agents that buy on behalf of customers should operate openly. Its Conditions of Use now require agents to identify themselves in every request. Meanwhile, Meta has made Shopify one of Muse’s connected services. Big retailers are choosing sides, and small businesses will have to decide too.

For most small and medium businesses, being easy for a legitimate agent to use is an opportunity. An agent that can read your prices, stock and policies, and complete your booking form or checkout, can send you a customer. One that can’t will pick a competitor. A quick check of your site:

• Products, services, prices, availability, shipping and returns are published as structured data (a standard format that software can read without guessing).
• Reviews are visible and machine-readable. We built our Shopify app, Review Optimiza, to sync a store’s latest reviews into structured data for exactly this reason.
• Booking forms and checkout work without tricks that only a human can get past, while still protecting against fraud.
• Your policy on AI agents is a deliberate choice, and your security settings match it.

If you sell on Shopify, our guide to getting your products recommended in AI shopping covers the catalog side in more detail. Much of this is the same work that gets you found in Google and recommended in AI answers.

When an off-the-shelf agent isn’t enough

General agents are good at work that looks the same in every business: email, calendars, research and common apps. They struggle when the job depends on your own systems, such as a custom job tracker, an inventory spreadsheet with its own logic, or software with no connector at all.

That’s where a purpose-built AI tool, connected to your data with the right permissions, tends to win. For a health and wellness business, we built a custom AI-powered ERP application that gave them a single source of truth and saved roughly 30 hours a week of manual tracking. The AI worked because it was designed around how that business already ran.

Always-on agents will get cheaper and more capable over the next year, and the businesses that benefit most will be the ones that set clear goals, kept tight access and made their own sites easy for agents to use. If you’d like help choosing where an AI worker fits, building one around your own data, or getting your store ready for the agents your customers are already using, get in touch.