sdecoret - stock.adobe.com

Meta expands Muse to small businesses

The move comes as agentic AI plays an increasingly big part in business operations, sparking oversight concerns.

Meta is rolling out a version of its AI agent, Muse, for small businesses, offering access to business accounts and third-party tools, including QuickBooks, Shopify and Slack.

First launched on Sept. 8, Muse reached 5 million downloads in North America in three weeks, displacing ChatGPT at the top of the U.S. App Store charts, where it still sits at number one as of today, despite emerging security concerns.

The expansion for SMBs, revealed on Tuesday, comes a day after OpenAI unveiled its Muse response: Dots, an always-on AI agent designed to work autonomously across more than 4,000 connected apps. OpenAI said Dots can pursue goals in the background, work through tools including Slack and Teams, and take actions on users' behalf, subject to permissions and approval controls.

The developments point to a broader push by AI companies to move beyond chatbots and toward agents capable of carrying out multi-step tasks across the software people use for work.

Muse for small businesses

Meta is positioning its new agent as just that -- a tool for users to hand off multi-step jobs across the software they already use. Rather than employ the agent to simply ask questions, users can give Muse broader goals, such as running a business or finding new customers.

Business owners can also now connect the agent to Facebook and Instagram business accounts, as well as Meta advertising accounts, to access business analytics and customer interactions.

Mark Beccue, an analyst at Omdia, a division of Informa TechTarget, said the expansion reflects growing interest in using agents to help small businesses, but questioned whether Meta can displace more established players.

“It shows there is momentum and interest from AI players to help SMBs,” he said. “However, I’m skeptical as to why SMBs would choose Muse to connect their business systems when there are more proven cross-systems players in the space that they may already work with, particularly Microsoft and Google.”

Beccue said small businesses could instead turn to prebuilt agents from platforms such as Google’s, while Muse may have a stronger fit among consumer-facing online retailers already operating in Meta’s ecosystem.

“When it comes to their agent play, SMBs will likely look to prebuilt agents from Gemini Enterprise, for example,” he said. “Muse might make sense for SMBs that are consumer-oriented retail online businesses.”

The question of trust may also play out differently for smaller companies than for larger enterprises. While connecting agents to operational, financial and customer data introduces obvious risks, Beccue said SMBs may be less concerned about handing data to agents than larger organizations.

“I think SMBs have fewer issues with trusting agents with their data than bigger businesses,” he said. “I think they’re more concerned with cost-effective solutions that require fewer resources.”

Safeguarding against increased autonomy

Despite this potential payoff, Wayne Hankins, an analyst at Gartner, said the need for governance should be a key consideration for businesses of any size that embrace increasingly autonomous systems.

“The biggest mistake executives can make is treating agent security as solely a technology or cybersecurity issue,” he said. “Agentic AI is ultimately a governance challenge.”

Hankins said organizations should establish clear boundaries around the data, tools, systems and decisions an agent can access before deployment, using controls such as least privilege, segmentation and runtime authorization. For small businesses in particular, the question becomes how much access an owner is prepared to grant an agent, and what controls remain in place once the agent has that access.

Hankins highlighted human approval for high-risk actions as an important safeguard as agents become more independent.

“As agent autonomy increases, organizations should implement human approval workflows for sensitive actions,” Hankins said.

The concerns are not entirely theoretical. Meta's consumer Muse has already faced scrutiny over its behavior, with The Guardian reporting this week that the agent provided a Facebook Marketplace buyer with a seller's home address and arranged a meeting without the seller's knowledge. Meta said it was investigating the incident.

That incident highlights the distinction between an AI assistant that generates information and an agent that can act on information across external systems.

Comprehensive logging, observability, audit trails and runtime monitoring are also needed to understand what agents are doing and ensure their actions remain within approved boundaries, Hankins said.

“Recent agent incidents reinforce that attackers continue to succeed through familiar weaknesses such as excessive privileges, poor segmentation, exposed secrets, weak identity controls, and unrestricted connectivity,” he said. “AI is accelerating the exploitation of these weaknesses, not replacing the need to address them.”

Scarlett Evans is a freelance writer with a focus on robotics and emerging technologies. Previously, she was assistant editor at IoT World Today, where she specialized in robotics and smart city technologies. Scarlett also has a background in the mining and resources sector, with experience at Mine Australia, Mine Technology and Power Technology.