The Move
On September 28, 2026, Meta Platforms announced a new business unit: the Meta Enterprise Platform. Led by CJ Desai, former CEO of MongoDB, the initiative aims to bring Meta's full AI stack, including Muse, Meta Business Agent, Muse Code, and the Muse API, to enterprises and developers.
Desai will hold the title of Chief Enterprise Platform Officer. He joins Meta after an 11-month stint as MongoDB's CEO. MongoDB stock fell more than 18% on the news. Dev Ittycheria, who led MongoDB for 11 years before Desai took the helm, returned as interim CEO.
Meta has shared limited details about the new business. The initial stated goal is to broaden the adoption of its Muse agent among enterprises and developers. But the broader ambitions are significant. According to Reuters and SiliconANGLE reporting on September 28, 2026, Meta may also sell excess data center capacity to other companies, a move Bloomberg and CNBC reported in July 2026.
What Meta Is Actually Selling
The Meta Enterprise Platform encompasses several products:
Muse Agent. Meta's consumer AI agent, which has reached 3.4 million downloads since launch. The enterprise version would bring the same agent capabilities to business workflows.
Muse Code. A programming agent launched in August 2026. Its distinguishing feature is an agent fan-out mechanism that splits complex programming tasks among multiple subagents, significantly speeding up processing. This puts it in direct competition with GitHub Copilot and Anthropic's Claude Code.
Meta Business Agent. A chatbot designed to field customer inquiries via WhatsApp. It answers common product questions, generates buying recommendations, and solves technical issues. This is currently the simplest product in the enterprise portfolio.
Muse API. Gives developers access to the large language models that power Meta's agents. Meta debuted its newest LLM, Muse Spark 1.3, earlier in September 2026. According to Meta, it outperformed GPT-5.6 Sol on AutomationBench and scored higher on five coding benchmarks. It uses approximately 25% fewer tokens than its predecessor for coding tasks.
The Security Roadmap
Meta's stated security approach is noteworthy because it is mostly forward-looking promises rather than shipped features. According to SiliconANGLE reporting on September 28, 2026:
- Muse Confidential VM, a feature that isolates Muse instances so completely that even Meta cannot access their contents, will roll out later in 2026.
- Meta plans to share a continuous audit that will enable users to verify the isolation technology works as intended.
- The company suggests it may add integrations with third-party cybersecurity tools, similar to how Anthropic offers connectors that send Claude usage logs to external breach detection platforms.
These are promises, not products. Enterprises evaluating Meta's enterprise AI stack should weight them accordingly.
The Enterprise Track Record Problem
Meta's history in enterprise software is not reassuring. The company has repeatedly launched enterprise products, built adoption, and then shut them down.
| Product | Launched | Shut Down | Users at Shutdown |
|---|---|---|---|
| Workplace (enterprise social) | 2016 | 2026 (winding down, gone by May 2026) | 7 million paid users |
| Horizon Workrooms (VR collaboration) | 2021 | February 2026 | Not disclosed |
| Meta Business Suite (original SMB tools) | Rolled into other products | Multiple iterations | N/A |
Workplace is the most striking example. Meta built a product that reached 7 million paid users and still killed it. Enterprises that adopted Workplace invested in integration, training, and workflow changes, only to see the product sunset. The Meta Business Agent product is currently a WhatsApp chatbot. The enterprise AI platform it belongs to was announced with a poached CEO and a product roadmap that includes no pricing, no admin controls, no compliance certifications, and no security features that are actually shipped.
The Muse Security Context
Even the consumer Muse agent has had security issues. A zero-day vulnerability in the Muse Mac app was patched within a day of discovery. VentureBeat reported that enterprise security teams still lack central visibility into what the Muse agent can access on a user's device. For a company that wants to sell enterprise AI, this is a credibility gap.
Compare this to the security posture of competitors. Anthropic operates Claude Managed Agents with security boundaries between the agent loop and execution sandboxes, and has integrated with NVIDIA's OpenShell for hardware-level enforcement. OpenAI, despite its own safety controversies with the cancelled Astra 6.1 model, has the OpenAI Agents API and Managed Agents with documented security boundaries. Google offers Vertex AI Agent Builder with IAM-based access controls.
Meta's enterprise security offering is a roadmap. Its competitors' security offerings are products.
The Strategic Logic
Despite the risks, Meta's enterprise AI push has underlying strategic logic. Meta has built substantial AI infrastructure for its consumer products. Its LLMs, Muse Spark 1.3, are reportedly competitive with or superior to GPT-5.6 Sol on certain benchmarks. The company has massive compute capacity from its advertising business. Entering enterprise AI is a way to monetize that infrastructure beyond advertising.
Hiring CJ Desai signals seriousness. Desai ran MongoDB, a public company with a market cap in the billions, and built a reputation in enterprise software. The fact that Meta created a new C-level role, Chief Enterprise Platform Officer, rather than burying the initiative under an existing executive, suggests organizational commitment.
But the question is not whether Meta is serious. The question is whether Meta is patient. Enterprise software requires sustained investment, multi-year sales cycles, compliance certifications, and the willingness to maintain products that generate less revenue per user than advertising. Meta's DNA is consumer and advertising. Its enterprise history suggests that when the patience runs out, the products get killed, regardless of how many users depend on them.
What Enterprises Should Ask
For any enterprise considering Meta's AI platform, the due diligence checklist should include:
1. What happens if Meta shuts this down? Meta has shut down two enterprise products in 2026 alone. What is your exit strategy if the Meta Enterprise Platform is sunset in 18 months? Can you migrate your agents, data, and integrations to another platform?
2. What security features ship on day one? Muse Confidential VM is a roadmap item. What security controls are available today? How do they compare to Anthropic's Claude Managed Agents, OpenAI's Managed Agents, or Google's Vertex AI Agent Builder?
3. What are the admin and compliance controls? Meta has announced no enterprise admin console, no SSO integration, no compliance certifications, no data residency controls, and no audit logging capabilities. When will these exist?
4. What is the pricing model? No enterprise pricing has been announced. Is it per-seat, per-API-call, per-agent, or something else? How does it compare to existing enterprise AI platforms?
5. How does Meta handle agent identity and access management? NVIDIA's Open Agent Safety Platform launched the same day with 100+ partners and hardware-level enforcement. Meta's Muse security roadmap is a promise. Which one would you bet your data on?
The Bottom Line
Meta has the resources, the talent, and the models to build a credible enterprise AI platform. CJ Desai is a serious hire. Muse Spark 1.3's benchmark performance suggests the underlying technology is real. But Meta's enterprise track record, its history of killing products that reached millions of paying users, and its current security roadmap (which consists mostly of future promises) mean that any enterprise considering the Meta Enterprise Platform should approach with structured skepticism.
The company that killed Workplace, a product with 7 million paid users, wants you to trust it with your enterprise AI infrastructure. Maybe this time will be different. But enterprises should plan for the possibility that it will not be.
Evaluating enterprise AI agent platforms? Contact Vivid for a structured comparison of Meta, OpenAI, Anthropic, and Google for your business.
Read next: The AI Shopping Disconnect: 61 Million Consumers Use AI to Shop, 89% of Merchants Aren't Ready | From Storefronts to Agent Endpoints: How Brands Are Distributing Themselves via AI | NVIDIA's Open Agent Safety Platform: Silicon Cages for Rogue AI Agents | Microsoft Autopilot: The 24/7 Cloud Agent That Works While You Sleep
