Meta's Muse agent connects to small firms' books and ads, and OpenAI's Dots keep working in Slack

Muse arrived earlier this month as a personal agent that shops, books travel and sends email. The version for small businesses connects it to QuickBooks, Stripe and Shopify and more than a dozen other work tools, plus the owner's Facebook Page, Instagram analytics and Meta ad account. It is free with usage limits, and paid plans buy more use.

Each of those systems answers one question well. The questions an owner really asks cross several of them at once. For example, a shop that sells kitchenware online could ask which of last month's Instagram campaigns brought customers who paid and kept what they bought. Muse can take the ad spend from the Meta account, the orders from Shopify, the refunds from Stripe and the product costs from QuickBooks, then propose moving money away from the campaign that lost it. By hand, that means exporting four files into a spreadsheet and matching them line by line, about half a day every month. Muse will not post, message or buy anything without the owner's approval, so the decision stays with a person.

The agent that reads the books is built by the company that sells the ads. So when Muse suggests where the next $1,000 of marketing should go, the seller of the ad space is giving the advice. A careful owner treats it like a proposal from a media salesperson and checks it against the margins in QuickBooks before approving.

A task that took a lawyer ten hours can take two with AI tools. Last year, only 6% of law firms passed those savings on to clients. That is harder to sustain now, because clients' own staff use the same tools and can draft much of the work themselves.

So law firms, agencies and consultancies are cutting prices, selling subscriptions instead of hours, and offering to check drafts that the client's own AI wrote. The outside firm is paid less for writing the first draft and more for standing behind the final one.

Each Dot runs on its own cloud computer, connects to more than 4,000 apps, and carries a project from ChatGPT into a Slack or Teams thread without a new briefing. ChatGPT could already spend hours on one assignment. A Dot keeps watching after the assignment is done: a purchasing team could have one follow supplier delivery emails, update the shipment tracker as they arrive, and message the buyer in Teams when a date slips.

An agent that acts while nobody is watching needs its limits set in advance. Dots accept rules about when they must ask for approval, and those rules deserve care: OpenAI has just shelved a more capable model because it crossed its permission limits during testing.

The notetaker, on Mac since August and on Windows since mid-September, records a meeting from the computer's own audio, so no bot appears in the participant list on Zoom, Teams or Google Meet. A visible bot shows everyone that a recorder is present. Here only the person recording knows, and Wispr's own page warns that participants may not get an automatic notification. Telling them becomes the user's job alone. Otter.ai, a rival notetaker, is already in court over recording consent. A simple team rule covers it: whoever records says so when the call begins.

Space is a shared area in ChatGPT where a team and its AI agents edit the same page at the same time. Collaborative slides follow in the coming weeks, and they export to PowerPoint and Google Slides.

That export shows what kind of rival this is for now. The draft is made in ChatGPT, and the finished file still goes out in Microsoft's or Google's format, because that is what clients and boards open. At the next Microsoft 365 renewal, some staff who start their work in ChatGPT may need only a lighter Office licence to finish it.

From October 30, the plan's included use of ChatGPT Work and Codex falls from 20 times the Plus allowance to 10 times, and the price stays $200 a month. Existing subscribers keep the old allowance until then.

OpenAI now sizes the allowance by what the same work would cost at its API prices, so the flat fee behaves like a prepaid budget. In a team, the heaviest users reach it first, and the choice between making them wait and moving them to the new $500 plan sets the real cost.