Meta's Muse, and where an AI agent should sit in your business

On 29 September Meta launched Muse for Small Business, a business edition of its Muse personal AI agent. If you run your marketing through Facebook and Instagram, you'll probably hear about it soon, and you may wonder how it fits with the AI you already have in Allwis.

The short answer: they do different jobs, and the useful question isn't which one to use. It's which job each one should do.

What Muse is

Muse is an agent, not a chatbot. You give it a goal, such as finding new customers, getting the inbox under control or working out why sales dipped, and it works through the steps across the apps you've connected.

Meta lists connections to Facebook Pages, Instagram professional accounts and Meta ad accounts. It also connects to a set of office tools: Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, Intuit QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe and Zoom, plus custom connectors. Meta's examples include analysing sales data, managing email and calendar, tuning ad campaigns and reviewing financial performance and expenses.

Two details matter more than the feature list:

• Meta's rule is that "nothing publishes, sends, or spends without your approval." Muse can draft and flag things on its own, but it asks before it acts.
• It is available in the US and Canada. Australian and New Zealand businesses can't use it yet.

Where an agent sits changes what it knows

Muse works through connectors. It sees what each connected app chooses to show it, and it acts by asking that app to do something. That's a reasonable design for an assistant that spans a dozen tools, and it suits marketing well: your Page, your ads and your audience all live inside Meta anyway.

Your books are a different case. Whether you can afford a campaign, whether a customer who just messaged you is 60 days overdue, whether that bill has already been paid: those answers depend on records that have to agree with each other. An agent working from the outside has to reassemble that picture through an API, one call at a time, every time.

That's why the AI in Allwis lives inside the records rather than beside them. The Allwis Concierge turns a plain-English request into a plan you can read, edit and confirm before anything runs, and it works with the same permissions you have. Some examples:

• "Chase every overdue account" gives you the list to approve before any reminder is sent.
• "Acme paid INV-0142" records the receipt against the right invoice and names the customer and the amount before you confirm.
• "Invoice the Hermes job" raises the draft from every unbilled hour, telling you the hours and the total first.

There are also things it won't do. It never lodges, signs or sends a tax return.

Meta has landed on the same principle we started with: AI proposes. People approve. That's a good sign for everyone. An agent that spends money or sends messages without asking first isn't one a business can trust.

Using both

For an Allwis customer, the natural split is:

• Meta's side: the audience. Page posts, Instagram, ad campaigns and the conversations those create. Muse is built for that, once it reaches your country.
• Allwis: the business behind it. Invoices, bills, the bank, payroll, jobs and the customer record those conversations turn into.

Some of that bridge already exists in Allwis. PPC Sync brings your Meta Ads spend and performance into your reporting, so ad cost sits beside the revenue it produced. The Social Publishing add-on drafts posts for your Facebook Page, Instagram account and LinkedIn Page using only facts you already hold, and nothing goes out until a person approves it.

Muse doesn't connect to Allwis today, and Allwis isn't on Meta's connector list.

Three questions for any AI agent

Whether it's Muse, the next one or the one after that, ask the same three things before you give an agent access to your business:

1. What can it do without asking? Reading and drafting are cheap to undo. Sending, paying and posting are not.
2. Whose permissions does it use? An agent should never be able to see or change more than the person it's acting for.
3. Can you tell afterwards what it did, and which AI did it? In Allwis, every deal score, audit note, email summary and ticket triage carries a chip naming the provider and model that produced it, and it stays accurate after you change providers.

Where your data goes

Muse runs on Meta's models in Meta's cloud. For your marketing that's unavoidable, because the audience is already there. For client files, payroll and the ledger, you may want more say.

In Allwis, Anthropic Claude is the default and you can bring your own model, in the cloud or on your own hardware. You choose which provider handles which task. You can also switch on self-hosted-only mode, which keeps prompt data on infrastructure you control.

The short version

• Muse is an AI agent for your marketing and office apps, with an approve-before-it-acts rule. It's available in the US and Canada and not yet in Australia or New Zealand.
• Agents are most useful where the records they act on live. For your audience, that's Meta. For your money, it's your books.
• The rule that matters is the same everywhere: AI drafts, flags and proposes; a person decides.

Source: Meta, "The Future Is for Everyone: Muse for Small Business", 29 September 2026, about.fb.com, checked 5 October 2026. The connector list, the approval wording, availability (US and Canada) and the use cases are taken from this page.