Meta Platforms on September 28, 2026 set up Meta Enterprise Platform, a division meant to turn the models and agents it built for its own apps into products that other companies can buy and run. Chirantan "CJ" Desai, until then MongoDB's president and chief executive, will lead it as Chief Enterprise Platform Officer, reporting directly to Mark Zuckerberg.

In Short

Meta, which earns almost all of its money from ads, said on September 28, 2026 that it will also package its AI tools for companies to use in their own operations. It hired MongoDB's boss, CJ Desai, to run the effort, and the first products are the Muse assistant, the Business Agent that answers customers on WhatsApp, and two tools for developers. For the businesses that already buy Meta ads, this means Meta wants to be their software supplier as well, though it has not said what the products will cost or when anyone can buy them.

What Meta published

The news took the form of a Newsroom post dated September 28, filed under company news and tagged with a single word: AI. It holds two statements, one attributed to Zuckerberg as founder and chief executive, the other to Desai. According to Bloomberg, Zuckerberg also posted it on X the same Monday.

Zuckerberg opened with a premise rather than a product. "We believe superintelligence will create significant new opportunities for all people and businesses," he wrote. Meta, he continued, already serves billions of people and helps hundreds of millions of businesses reach customers. The new division is "the next major pillar of our business," set up "to help businesses use AI to grow and transform in new ways as well."

The claimed edge is compressed into a single sentence. Meta Enterprise Platform, according to Zuckerberg, will draw on "advanced models, leading agents, large-scale infrastructure, and years of working closely with many businesses," strengths he said few other companies possess. The first task is narrower. Meta will initially concentrate on bringing its "full technology stack" to businesses and developers, and four products are named: the Muse agent, Meta Business Agent, the Muse API and Muse Code. The list ends with the words "and more".

Desai's statement moves from the company's history to its intentions. "Over the coming years, AI will fundamentally redefine how organizations of all sizes innovate, grow, serve customers, and run business operations," he wrote. The division, in his words, "will focus on turning its AI stack into products and services that companies can deploy for their own businesses." His stated goal is "to make Meta the place enterprises come to scale their businesses."

What is missing

Taken together, the two statements run to a little over 400 words. They name no customers, no prices, no availability dates and no markets. There is no figure for the division's staff, budget or revenue ambition, and no indication of whether Meta will sell through its existing Meta for Business organisation, through partners, or through a separate sales force under Desai. Nothing specific is said about how customer data handled by the enterprise products will be treated in relation to Meta's advertising systems, beyond Desai's general assurance on security and privacy.

That silence is conspicuous because Meta's consumer AI already runs under two different data rules. Since December 16, 2025, conversations with Meta AI have fed content and ad personalisation across Meta's apps, a change covering more than 1 billion monthly users. Muse, by contrast, was published with a commitment that its conversations and the data on each person's virtual machine stay out of Meta's ad systems. Which of those rules would govern a company deploying Meta's agents has not been stated.

The four products

Muse agent

Muse is the newest of the four and the one with the least obvious enterprise use. Meta released it on September 8, 2026 as a consumer agent that runs on its own virtual machine, opens its own browser and completes purchases with a one-time card, available in the United States only, on iOS, Android and muse.ai. Each person's agent runs on a dedicated Muse Secure VM that stores credentials for connected services, and a separate program, Sentinel, has to approve anything the agent sends to the internet. A Muse Confidential VM, encrypting the whole machine under a key held only by the user, is due later in 2026.

Desai leaned on that design. "As with Muse, security and privacy are built into Meta's enterprise products from the outset," he wrote. How a personal agent built around one person's goals becomes something a company deploys for staff or customers is not explained, nor who inside a business would control it, nor whether the one-user, one-machine model carries over.

The consumer rollout has been contested. In the week to September 27, Amazon blocked Muse from its store after Meta declined to exclude Amazon from the agent's reach, and complained that the agent captured shoppers' credentials without identifying itself, as PPC Land's weekly edition recorded. The same edition cited Digiday figures of 730,000 US downloads in five days, with revenue limited so far to $20 and $100 monthly tiers plus merchant fees, and noted an AdExchanger report that some Muse phone calls had been placed by people in a call centre.

Meta Business Agent

The second product sits far closer to Meta's existing commercial relationships. Meta Business Agent was opened to businesses of all sizes worldwide on June 3, 2026, at the Conversations event in London, handling customer conversations across WhatsApp, Messenger and Instagram, with a free entry tier and paid subscriptions planned. More than 1 million businesses had run a version of it on WhatsApp and Messenger before that date, according to Meta.

An enterprise tier already exists in some form. In June, Meta described a separate Meta Business Agent Platform for larger organisations, connecting to hundreds of third-party systems, with Shopify and Zendesk named. On the second-quarter earnings call on July 29, chief financial officer Susan Li described the platform as infrastructure for enterprises to build, customise and deploy agents at scale on WhatsApp, with controls, guardrails and measurement built in, and said it had been introduced earlier that month. The two accounts date the platform differently. The September post does not clarify which stage of the product the new division inherits.

Scale figures come from Meta itself. Zuckerberg told analysts that more than 1 million businesses were using Business Agents each week. Li cited Movida, a Brazilian car rental company with nearly 400 locations, which reported a 44% rise in daily WhatsApp bookings over a one-month period compared with a year earlier, and 85% of conversations resolved without human involvement. Neither figure has been independently verified.

Muse API and Muse Code

The last two names arrive without definitions. The nearest documented product is the Meta Model API, which opened in public preview on July 9, 2026, when Muse Spark 1.1 reached outside developers with a 1 million token context window. Whether "Muse API" is a new name for that interface, a commercial tier of it or something separate, the post does not say. Muse Code receives no description at all - no model, no pricing, no access terms - and PPC Land has not previously reported on a Meta product under that name.

Desai's record

Zuckerberg's statement supplies the career outline. Desai joins from MongoDB, where he was chief executive and president. Before that he led product and engineering at Cloudflare, and he spent "nearly eight years at ServiceNow, including as President and COO." Zuckerberg described him as "an experienced enterprise leader with a track record of building full-stack software and delivering results in AI, infrastructure, business applications, and security," with "an extensive industry network that we look forward to partnering with."

His spell at MongoDB was brief. The database company said on November 3, 2025 that its board had appointed him president and chief executive with effect from November 10, succeeding Dev Ittycheria, who had run MongoDB for 11 years. According to CNBC, MongoDB's board has now named Ittycheria interim president and chief executive, and a person familiar with the matter told the broadcaster that Meta's offer was large enough for Desai to give up his MongoDB pay package and severance. MongoDB shares fell sharply on September 28, according to CNBC and Bloomberg, though the size of the decline quoted by different outlets varied through the session.

None of the three companies on that record sells advertising. MongoDB, Cloudflare and ServiceNow sell databases, network security and workflow software to corporate technology departments, a different buyer from the marketing teams that supply nearly all of Meta's revenue. The reporting line reinforces the separation. Desai answers to Zuckerberg - not to the advertising organisation, and not to the Meta Superintelligence Labs teams that build the Muse models.

The money behind the pillar

Meta's second-quarter results show how concentrated the business the new division is meant to broaden remains. Total revenue was $60.80 billion, of which advertising contributed $59.36 billion, leaving less than 3% from everything else. The Family of Apps "other revenue" line, which includes subscriptions and WhatsApp paid messaging, passed $1 billion in a quarter for the first time, up 73% year over year.

Spending has grown far faster than those alternative revenue lines. Capital expenditures, including principal payments on finance leases, reached $31.08 billion in the quarter against $17.01 billion a year earlier, while free cash flow fell to $784 million from $8.55 billion. Full-year 2026 capital expenditure guidance started the year at $115 billion to $135 billion, was raised to $125 billion to $145 billion after the first quarter, and was narrowed to $130 billion to $145 billion in July. Li declined to give an outlook for 2027. On the same call, Zuckerberg said Meta was fielding a large number of offers for its compute but regarded its internal uses as more valuable - context worth setting beside a division that lists "large-scale infrastructure" among its selling points.

The enterprise idea did not appear from nowhere this week. Zuckerberg's statement in the July earnings release already counted new enterprise opportunities among the things AI was making possible for Meta. The company cut about 8,000 roles in May 2026 and reported 75,472 employees at June 30. Whether the new division will be staffed through hiring or through reassignment, the September post does not say.

From personal superintelligence to enterprise

The vocabulary in the September post has a lineage. On July 30, 2025, Zuckerberg set out a vision of personal superintelligence for everyone. On August 10, 2026, he published a 6,542-word letter that, by PPC Land's count, used the phrase "balance of power" 23 times and did not mention advertising once. That letter drew a line between Meta and other AI laboratories, which Zuckerberg described as focused on building AI for companies, governments or other institutions, and warned that the balance of power would tilt towards larger institutions if those laboratories led.

Seven weeks later, Meta has a division whose stated purpose is building AI products for companies. The September post frames the move around "all people and businesses," and Desai writes of organisations "of all sizes." Is this a change of strategy, or an extension of it? The post does not engage with the earlier contrast, and nothing in it explains how an enterprise business fits the argument made in August.

What it means for advertisers and agencies

Desai's statement points to the likely first customers. Meta, he wrote, has "a proven track record of helping millions of advertisers and hundreds of millions of businesses scale." Many of the companies the division will approach already buy Meta media and already hand much of that buying to Meta's automation. Advantage+ passed an annual revenue run rate above $75 billion in the second quarter, and more than 9 million small businesses were using at least one of Meta's AI ad creative tools. Earlier in the year, business AI conversations on WhatsApp and Messenger rose from 1 million to more than 10 million a week.

Meta's advertising system has also been opened to software throughout 2026. On April 29, Meta gave agents from Anthropic and OpenAI write access to ad accounts through connectors built on a Model Context Protocol server, and on July 16 it opened that server to any developer holding a Meta app. Google took the opposite line on August 31, when it prohibited programmatic proxies, including MCP servers that only re-expose Google Ads functions. Meta had earlier bought Manus, a Singapore-based agent developer, in a deal disclosed on December 29, 2025, and put its agent inside Ads Manager. By then, according to Meta, Manus had processed more than 147 trillion tokens and created more than 80 million virtual computers.

Set beside the enterprise announcement, those pieces describe one supplier across several layers of a brand's operation: media buying through Advantage+, customer conversations through Business Agent, campaign management through agents connected to Meta's ads server and, if the Muse API follows the Meta Model API, model access for in-house tools. For agencies, and for vendors that built agentic AI and chatbot products on top of WhatsApp and Meta's advertising interfaces, the platform owner is now a declared seller in the same categories.

Regulators are already examining that layer. On June 9, 2026, the European Commission ordered Meta, through interim measures, to restore free WhatsApp Business API access for third-party general-purpose AI assistants on the terms that applied before October 15, 2025, in case AT.41034. That order governs access for rival assistants rather than Meta's own products, yet it sets the conditions under which competitors reach businesses on WhatsApp while Meta sells its own agent there. France's competition authority concluded on July 17, 2026 that OpenAI, Google and Anthropic held more than 84% of the global AI agent market, citing Sensor Tower data for May, and cited Meta AI's presence inside WhatsApp among the default placements that make it hard for newcomers to reach users.

What remains open

The September 28 post establishes a leader, a reporting line and a product list. It does not establish a price list, a timetable, contractual data terms or a geographic footprint, and Muse itself remains confined to the United States. The questions that will decide the division's weight for advertisers are therefore unanswered: whether data from enterprise deployments can inform ad delivery, whether Business Agent pricing changes as the product moves under the new division, and whether Meta sells its enterprise products alongside media or apart from it. Meta gave no date for further detail.

Timeline

Summary

Who: Meta Platforms, through statements from founder and chief executive Mark Zuckerberg and from Chirantan "CJ" Desai, the former president and chief executive of MongoDB, now Chief Enterprise Platform Officer reporting directly to Zuckerberg. MongoDB named Dev Ittycheria interim chief executive.

What: A new business division, Meta Enterprise Platform, intended to turn Meta's models, agents and infrastructure into products companies can deploy themselves. The initial stack named is the Muse agent, Meta Business Agent, the Muse API and Muse Code. No prices, dates, markets, customers or data terms were given.

When: Published on September 28, 2026, three weeks after Muse reached US users on September 8 and two months after Meta's second-quarter results on July 29.

Where: Meta's Newsroom, with a parallel post by Zuckerberg on X. No geographic availability was stated; Muse, one of the four named products, is available only in the United States.

Why: Meta drew more than 97% of second-quarter revenue from advertising while guiding to between $130 billion and $145 billion of capital expenditure for 2026. The division is its attempt to sell the AI stack to businesses directly, many of which already buy its advertising, raising unanswered questions about pricing, data separation and competition on WhatsApp.