By Ai Ti

Edited by Shen Xiao

Meta has just unveiled its new business unit, Meta Enterprise Platform, with plans to bundle offerings like the Muse intelligent agent, Meta Business Agent, Muse API, and Muse Code for sale to enterprises and developers.

Simply put, this marks the enterprise service iteration of Muse, the personal AI assistant that has recently surged in popularity. Leveraging Muse's consumer market allure, Meta aims to seize significant opportunities in the B2B arena.

Almost concurrently, upon hearing this news, the CEO of a listed company, who had been at the helm for less than a year, resigned to join Meta and promote its enterprise version of Muse to Zuckerberg.

On September 28, MongoDB made a sudden announcement: its CEO and President, Chirantan “CJ” Desai, who had been with the company for less than a year, would be departing immediately to join Meta as Chief Enterprise Platform Officer, reporting directly to Zuckerberg. Following this news, MongoDB's stock price plummeted nearly 18% during intraday trading.

MongoDB had originally scheduled an Investor Day for September 29 to unveil its long-term strategy, AI products, and growth prospects to the market. However, on the eve of Investor Day, the abrupt loss of its CEO, who had been in office for less than a year, necessitated the temporary return of former CEO Dev Ittycheria.

As a listed company with a market capitalization of $20 billion, MongoDB is also heavily investing in AI data infrastructure. Its Atlas database, along with its search, vector retrieval, and real-time data capabilities, forms the bedrock for enterprises developing AI applications.

MongoDB's Q2 financial report revealed revenue of $771.8 million, a 30% year-over-year increase, with Atlas-related revenue reaching $565.9 million, up approximately 29%. The company also raised its full-year revenue guidance to a range of $2.99 billion to $3.03 billion.

Desai's decision to leave MongoDB for Meta at this juncture underscores the immense allure of Meta's AI ecosystem.

Recent data from Sensor Tower indicates that Muse boasts around 1.8 million daily active users (DAUs) but generates daily revenue of approximately $15,000. Based on subscription pricing, the penetration rate of paying users hovers around 1%–2%. While DAUs continue to rise, daily revenue has peaked and is now on the decline.

This divergence may reflect short-term volatility, but it also highlights Muse's current primary risk: not its AI capabilities, but the low frequency of use cases.

Typical Muse use cases involve AI agents handling personal tasks such as canceling subscriptions, organizing past-year bills, comparing insurance plans, processing tax returns, booking flights, or canceling trips. Users are willing to delegate these tasks to an agent because it can save them significant time and even money.

However, under AI's efficient processing, personal tasks often require months of accumulation to justify a single Muse session. This creates a classic contradiction: high single-use value but extremely low frequency of use.

This differs from ChatGPT's business model. ChatGPT supports a wide range of activities, including writing, searching, programming, learning, and file summarization, with users generating new demands daily. While the single-use value may not be as high, the combination of “high-frequency demand × high-frequency use” sustains subscriptions.

Muse's challenge lies in transforming one-time tasks into long-term management.

For instance, product design is evolving to address this issue. Muse will remember user preferences and proactively offer suggestions based on past information. If a user mentions a friend's dietary restriction once, Muse can recall it when organizing a dinner. If a user saves a recipe on Instagram, Muse can further generate a shopping list.

This means Muse is no longer just a tool to be opened “when needed” but is striving to become an agent that continuously manages personal life.

At the Connect conference, Meta introduced retail connectors like Walmart, Best Buy, American Eagle, Sephora, Ulta, and Wayfair, along with Shop Pay and PayPal payment options. For travel, it integrated Expedia; for daily shopping, Instacart; and for work, Notion, Granola, GitHub, and Box.

The significance of these connectors and payments lies in transforming Muse's business model, opening up revenue opportunities beyond subscriptions. Muse is evolving from a “high-value, one-time tool” into a “low-friction, continuously operating personal AI infrastructure.”

However, consumer products still face an inherent limit: how much users are willing to pay.

Apple and Google are poised to launch more system-level agents. Apple owns a vast ecosystem, including the iPhone, Mac, Apple Watch, AirPods, the App Store, and numerous native apps; Google controls Chrome, Gmail, Maps, YouTube, Android, and Google Drive.

From understanding users to invoking services and completing payments, these companies possess more comprehensive entry points and system privileges than ordinary AI apps.

Muse provides a crucial direction, but it is far from the final destination.

The bigger issue lies within Meta itself.

In Q2 2026, Meta reported revenue of $60.8 billion, a 28% year-over-year increase; total costs and expenses reached $42 billion, up 55%; capital expenditures for the period were $31.08 billion. The company expects full-year capital expenditures to range from $130 billion to $145 billion.

These funds are being invested in data centers, chips, model training, and infrastructure.

AI has already enhanced Meta's recommendations, advertising, and user engagement, but these gains still do not fully justify the escalating investments. Zuckerberg needs a larger, higher-priced revenue stream that directly appears on bills.

Thus, Meta is beginning to sell the capabilities behind Muse to enterprises.

During the July earnings call, Zuckerberg mentioned that Meta's enterprise opportunities include APIs, business agents, developer tools, and even direct sales of computing power. He also acknowledged that serving enterprises represents “a different muscle” that Meta is not yet accustomed to.

On September 28, Meta established Meta Enterprise Platform, with plans to bundle technologies like Muse, Meta Business Agent, Muse API, and Muse Code into products that enterprises and developers can purchase.

This creates a clear division of labor between Muse and the Enterprise Platform: Muse addresses usage frequency and subscription issues for consumer AI; the Enterprise Platform tackles Meta's AI investment challenges related to average deal size and ROI.

The former aims for daily user engagement; the latter seeks long-term enterprise payments.

Models form the foundation, agents serve as the entry point, connectors act as the execution network, computing power provides the infrastructure, and enterprise contracts generate revenue.

Large model companies are now on a recruitment spree for key talent.

Initially, the industry competed for chief scientists, model architects, and computing power experts. Today, as model capabilities mature into products, the ability to sell these products at a premium has become paramount.

This is why Meta recruited Desai from MongoDB.

Desai spent nearly eight years at ServiceNow, serving as President and Chief Operating Officer, helping drive annualized revenue from $1.5 billion to over $10 billion. He later oversaw product and engineering at Cloudflare before becoming MongoDB CEO in November 2025, succeeding Dev Ittycheria.

His career exemplifies a To B product growth model: understanding complex products, building customer relationships, driving sales organizations, completing large-scale deployments, and transforming products into long-term contracts.

These are precisely the capabilities Meta has historically lacked.

The mission of Meta Enterprise Platform is not to create another chatbot but to repackage models, agents, APIs, computing power, and business tools into products that enterprises can procure.

This is where talent like Desai adds significant value.

He can help Meta understand why enterprises buy or reject products and what kind of AI offerings can transition from one-time projects to long-term contracts.

OpenAI's frequent adjustments to its enterprise revenue leadership in under a year also underscore this point.

In August 2026, OpenAI appointed Dali Rajic as its new Chief Revenue Officer. Rajic had previously served as President and COO of Wiz, President of Zscaler, and Chief Customer and Revenue Officer of AppDynamics. OpenAI's mandate for him was clear: build the “revenue operations system” needed for the next stage.

Such appointments represent a different organizational focus than recruiting scientists.

Scientists push the boundaries of capability; revenue leaders expand the boundaries of customers. The former determines what models can do; the latter determines what enterprises are willing to pay for.

The speed of this talent transformation is driven by changes in financial data.

In December 2025, OpenAI appointed Denise Dresser, former Slack CEO and Salesforce executive, as Chief Revenue Officer to oversee enterprise business and customer success. At the time, OpenAI disclosed having over 1 million enterprise clients, including Walmart, Morgan Stanley, Intuit, and Databricks.

By April 2026, OpenAI revealed that enterprise revenue accounted for over 40% of total revenue, with plans to approach consumer business revenue by year-end. Codex reached 3 million weekly active users, with APIs processing over 150 billion tokens per minute.

When enterprise revenue exceeds 40%, the company's top internal priority shifts from model deployment to sales team expansion, customer retention, usage growth, and maximizing revenue per customer.

Anthropic is also expanding its enterprise business along similar lines.

The company disclosed that since early 2026, enterprise subscriptions for Claude Code have grown approximately fourfold, with enterprise clients now contributing over half of Claude Code's total revenue.

After all, scientists determine the technical ceiling of AI model capabilities, while business operators control the profit-and-loss statement as this technology enters the commercial arena.