Meta Platforms Inc. is launching a new business unit that will provide artificial intelligence services to enterprises. 

The Meta Enterprise Platform, as it’s called, will be led by longtime technology executive CJ Desai. The company stated in a launch announcement today that he will hold the title of chief enterprise platform officer.

Desai is joining Meta following an 11-month stint as MongoDB Inc.’s chief executive. Shares of the database maker fell more than 18% today on the news of his departure. MongoDB said Dev Ittycheria, who had led the company for 11 years before Desai took the helm, has returned as interim CEO.

Meta has shared few details about its new AI business. It did, however, specify that the initial goal will be to broad the adoption of its Muse agent among enterprises and developers. The go-to-market push will also focus on several other products, notably Muse Code and Meta Business Agent.

Muse Code is a programming agent that Meta launched in August. One of the features that sets it apart from the consumer version of Muse is its so-called agent fan-out mechanism. The capability splits complex programming tasks among multiple subagents, an arrangement that significantly speeds up processing.

Meta Business Agent, the third offering that the company’s new AI business will promote, has a simpler feature set. It’s a chatbot designed to field customer inquiries via WhatsApp. Meta Business Agent can answer common product questions, generate buying recommendations and solve certain technical issues.

The company stated today that the Meta Enterprise Platform also encompasses its Muse application programming interface. The API gives developers access to the large language models that power its agents.

Meta debuted its newest LLM, Muse Spark 1.3, earlier this month. It outperformed GPT-5.6 Sol on AutomationBench, a benchmark that measures models’ ability to perform knowledge work. Muse Spark 1.3 also scored higher on five coding benchmarks.

Meta engineers prioritized cost-efficiency while training the model. According to the company, Muse Spark 1.3 uses about 25% fewer tokens than its predecessor when performing coding tasks. The improvement partly stems from the fact that the model makes more limited use of external tools.

“Meta Enterprise Platform will focus on turning its AI stack into products and services that companies can deploy for their own businesses,” Desai said in a statement. “As with Muse, security and privacy are built into Meta’s enterprise products from the outset.”

The company’s AI security roadmap includes a feature called Muse Confidential VM that will roll out later this year. It isolates Muse instances to such an extent that not even Meta can access their contents. For added measure, the company plans to share a “continuous audit” that will enable users to verify the technology works as intended.

Meta has a strong incentive to add more cybersecurity features down the line. Such capabilities would make Muse more appealing to risk-averse enterprises looking to mitigate the risk of AI hallucinations. Meta could, for example, add integrations with popular third-party cybersecurity tools. Anthropic PBC offers connectors that send Claude usage logs to external breach detection platforms for analysis. 

It’s believed that Meta’s enterprise AI portfolio will also include infrastructure services. In July, Bloomberg and CNBC reported that the Facebook parent intends to sell some of its excess data center capacity to other companies.

Photo: Wikipedia Commons