Meta Taps MongoDB's Desai to Lead Enterprise AI Push, But Llama's Role Remains Unclear

Meta unveiled plans Monday to establish a dedicated enterprise division centered on artificial intelligence, recruiting CJ Desai, who until recently led MongoDB, to oversee the initiative. The new unit, branded Meta Enterprise Platform, will adapt technologies developed for Meta's consumer applications and advertising business to serve corporate clients and engineering teams seeking to integrate these tools into their own infrastructure.

In a post on X, Mark Zuckerberg characterized the venture as a "next major pillar" of Meta's business, positioning enterprise AI alongside the company's existing advertising and consumer-facing operations. Desai assumes the title of Chief Enterprise Platform Officer and will answer directly to Zuckerberg. He departed MongoDB immediately, having assumed the CEO role just months earlier in November 2025. MongoDB subsequently appointed former CEO Dev Ittycheria as interim president and CEO.

The immediate question for the developer community centers on what capabilities this platform will provide. Zuckerberg outlined Meta's intention to deploy its "full technology stack" to enterprise clients and developers, beginning with the Muse agent, Meta Business Agent, Muse API, and Muse Code.

Meta introduced Muse this month as a consumer-oriented AI agent, while Meta Business Agent, which debuted in June, manages customer service interactions for businesses across Meta's ecosystem. Muse API and Muse Code represent the offerings most directly aimed at engineering teams, positioning Meta in direct competition with OpenAI, Anthropic, and Google in the market for agent and coding workflow solutions. Muse has already expanded beyond Meta's own platforms, with Shopify incorporating it across its merchant network even as Amazon declined to do so.

https://x.com/finkd/status/2104550609403695581?ref_src=twsrc%5Etfw

Enterprise pricing and availability remain unannounced

Meta withheld enterprise pricing, general availability timelines, and service-level agreements for Muse API and Muse Code during Monday's announcement, despite both products already being accessible to developers. Muse Code entered beta testing in August, and Meta began monetizing its Muse Spark model through its API in July, charging $1.25 per million input tokens and $4.25 per million output tokens.

When these terms do materialize, development teams should carefully evaluate them, particularly regarding how Meta manages model updates, since modifications to underlying models can disrupt production systems already in operation.

Llama's position in the enterprise strategy remains undefined

Llama, the model family Meta has spent considerable effort promoting to developers as an open-weights alternative for teams preferring to operate and customize models on their own systems, does not feature in Meta's public description of its new enterprise technology offerings. Meta has not clarified whether Llama will be incorporated into the Enterprise Platform. While Meta began directly charging developers for one of its proprietary models for the first time in July through Muse Spark, the company has also released open-weights versions of its smaller Muse Glimmer model and committed to an open-weights release of Muse Spark, making it premature to assess what this shift could mean for organizations currently deploying Llama in live environments.

Desai brings infrastructure and data expertise

During his tenure at MongoDB, Desai was already grappling with the infrastructure requirements for deploying agents at scale. In May, the company enhanced its data platform with persistent agent memory, automated embeddings, and additional AI-focused capabilities, with Desai asserting that the underlying data infrastructure, not merely the model itself, determines whether agents can successfully reach production. Meta confronts an analogous challenge: agents operating within enterprise environments require persistent context, real-time access to evolving datasets, and integration with the software tools employees already depend on. Comparable approaches are being pursued elsewhere, from Perplexity, whose agents constructed a database they lacked authorization to operate, to Microsoft, which equipped its Copilot agents with dedicated email, calendar, and organizational roles.

Desai's previous positions underscore this infrastructure focus. He directed product and engineering at Cloudflare, which has since pursued positioning itself as the foundational economic layer of the AI web, and spent nearly eight years at ServiceNow, ultimately rising to president and chief operating officer.

In remarks accompanying the announcement, Desai stated that Meta Enterprise Platform will concentrate on converting Meta's AI capabilities into deployable products and services for corporate use, emphasizing that security and privacy protections are embedded into Meta's enterprise offerings "from the outset." Although Meta has disclosed its security and safety methodology for Muse, Monday's announcement did not address data retention policies, whether customer data informs model training, tenant isolation mechanisms, identity management systems, or compliance certifications—information enterprise customers will require before granting Meta's agents access to sensitive internal data and systems.

Meta's existing relationships provide a foundation

Meta maintains established connections with numerous organizations it aims to serve through this initiative. Zuckerberg has highlighted the billions of individuals using Meta's services and the hundreds of millions of businesses operating on its platforms, many of which are small and midsize enterprises already leveraging Facebook, Instagram, and WhatsApp for marketing and customer engagement. Enterprise Platform could deepen these relationships by embedding Meta's AI capabilities directly into companies' internal operations.

Persuading engineering teams at larger organizations will present a greater challenge, given that their technical staff have invested the past several years building systems around models and platforms from OpenAI, Anthropic, Google, and competitors. Meta will need to demonstrate distinctive advantages that justify adding another platform to their technology stack or transitioning away from their current providers.

The platform remains partially opaque to potential users

At present, Meta Enterprise Platform functions more as a strategic vision than a fully realized product, despite certain components already being available to developers. A fundamental question persists unanswered: whether Meta's enterprise AI roadmap continues to feature Llama, or whether Muse signals a transition toward a proprietary platform model in which developers access Meta's most advanced agent technology exclusively through Meta's infrastructure, even as the company pledges open-weights availability for Muse Spark.