What Meta's business AI learns from a conversation
Meta now has a business unit for selling AI to companies. On September 28 it launched Meta Enterprise Platform, led by CJ Desai, previously CEO of MongoDB, who reports to Mark Zuckerberg. Desai’s announcement says AI will “fundamentally redefine how organizations of all sizes innovate, grow, serve customers, and run business operations.”
Snowflake, Databricks and Google have all made the case that an AI agent is only as useful as what it knows about your company. Meta makes the same case, and adds a key detail: the agent should learn from conversations.
What Meta built
The platform starts with products Meta already ships. The announcement leads with two of them, the Muse agent and Meta Business Agent.
Muse, launched September 8, is what Meta calls a personal AI agent. It works across the apps a person connects, sending email, booking travel and filling out forms, and Meta says it “remembers what matters to a person, so it can make suggestions unprompted and act on details that person only mentioned once.”
On September 29 Meta extended it with Muse for Small Business, which connects to tools including Shopify, Slack, Notion, QuickBooks and Stripe. Connect an owner’s Instagram, Facebook Pages and ad accounts, Meta says, and Muse “already understands your business: what you sell, what your brand sounds like, and what customers keep asking you about.” It flags emails that need a response and drafts replies, but Meta says Muse doesn’t publish, send or spend without the owner’s approval.
Meta Business Agent answers customers on a business’s behalf in WhatsApp Business, and has been available worldwide since June. It “learns automatically from your Facebook Page, past chats, and website,” gathers customer details, and alerts the owner when something needs their attention. The owner can choose which topics it hands over, and “When you take over, your AI learns from the conversation to improve future responses.”
So Muse learns about its owner and the tools the business uses, and Business Agent learns from the business’s conversations with its customers.
An order that changes in an email
Imagine a small catering company. A customer emails: “Can we move Friday’s lunch for 40 to Saturday? And we’ve moved offices, so it goes to 200 Elm Street now.”
Muse flags the email as needing a response and drafts a reply. The owner approves it, “Done, Saturday at noon at 200 Elm,” and the customer has an answer. Had the customer written on WhatsApp Business instead, Business Agent would have handed the chat to the owner and learned from the reply, so the next customer who asks to move an order gets a better answer.
Meanwhile the kitchen still has the lunch on Friday’s prep list, the driver has the old address, and the order in Shopify says Friday. Muse helped answer the customer, but the request also means facts that other people act on have to change, and the new facts are in one thread and in the owner’s head.
Muse could do more. It connects to Shopify and Slack, so it could update the order and tell the kitchen, and since it makes suggestions unprompted, it might even propose doing so. Muse is a personal agent, though. The suggestion goes to the owner, who has to approve it on a busy Thursday with the next customer already writing in. And because Muse is a personal agent, what it learns from the change is specific to the owner who uses it.
Learning from a conversation and recording what it changed
Both agents learn across many conversations. Business Agent answers better because customers ask similar things, and Muse becomes more helpful over time because it keeps a record of what matters to its owner. A change a customer sends in a single message is different. It comes up once, about one customer and one order, and some of the people who have to act on it may never have seen the thread.
Meta describes Muse as acting on “details that person only mentioned once,” which is the right goal. In the catering scene, the person who mentioned the detail was a customer, and the people who need it are the kitchen and the driver.
Desai named serving customers and running business operations as two of the things AI will change. The moved lunch needs both: someone answers the customer, and the kitchen and the driver change what they do on Saturday.
At Facebook, I led a product team building knowledge management products, for internal use and for external customers via Workplace, the version of Facebook that Meta sold to companies. A knowledge base only holds what someone remembered to write down. That step is why changes like the Saturday lunch go unrecorded. The customer asked for the change in their message, but unless someone moves that information into the system the delivery team works from, there’s a danger that the customer’s delivery will go to the wrong place on the wrong day.
How to automate it
Something reads the business’s messages on the team’s behalf, notices that this one asks to move an order and gives a new address, and proposes the updates in the system the team works from: a task on Saturday’s list for the kitchen, and the new address where the driver will look for it, each linked to the message it came from. Whoever on the team is closest to it confirms them, and from then on the people and AI assistants that read from those places get the same answer. Meta’s agents keep learning how to answer, but keeping the record is a separate job.
That’s what we’re building at this+that. DoBox reads the messages a company already gets across Gmail, Outlook, Slack, Microsoft Teams, and Google Chat, and turns them into tasks visible to everyone who needs to see them, then hands each to a workflow that handles it or to the person who should handle it. Our brain holds what the company knows, kept current today by workflows that write back to it as they run, with a person approving or editing each update.
Try it using your messages from the previous week. Pick a change a customer sent in a message, like a new date or a new address, and check whether everyone who needs to know about the change does. If not, see how our brain can help.