The Morning Download: Meta AI Agent Taps Booming Business Market

Good morning. Meta Platforms is accelerating efforts to move beyond the consumer market and tap booming business demand for AI.

On Wednesday, the company announced global availability of an artificial-intelligence agent for businesses on its WhatsApp, Instagram and Messenger services. Highlights from the Wall Street Journal story:

• The autonomous tool will be able to answer customer questions, book appointments and close sales, among other functions. In the future, Meta says its capabilities will include conducting market research, providing competitive intelligence and connecting with tools to manage businesses’ calendars.
• “As our models advance, your agent will take on more and eventually help you run your whole business,” Meta Chief Executive Mark Zuckerberg said during taped remarks at a London conference where the launch was announced.

The Journal’s Meghan Bobrowsky also reports exclusively that Meta has repeatedly delayed the release of its Muse Spark AI model API to developers, with no set date as of Tuesday. More on that below.

As the number of daily active users declines sequentially, Meta is looking beyond the consumer market to businesses as a source of revenue to support its AI investments. Its AI infrastructure build-out will consume up to $145 billion this year in capital expenditures, according to the Journal.

Meta has been testing the business agent in some countries in a limited capacity and said it already has one million businesses using it. It's also introducing a separate agent platform for larger companies that already use WhatsApp to manage their operations.

In past tech cycles, innovation has often shifted from the consumer market into the business market. Businesses have been key drivers of AI demand, given their willingness to pay for potential gains in both efficiency and growth.

Competition in the enterprise AI market is intensifying. The incredible surge in revenue at Anthropic reflects the model developer’s early emphasis on the enterprise market and OpenAI has sharpened its focus around the enterprise as well.

How Virgin Atlantic convinces employees to use corporate-sponsored AI. Richard Masters, vice president of data and AI at Virgin Atlantic, oversees the two functions for the British airline. Masters sat down with The Wall Street Journal Leadership Institute's Belle Lin this week to discuss the need for data and AI to work together, and how he helps employees choose between AI tools.

Below is lightly edited version of the conversation:

WSJLI: What does your role look like in practice?

Masters: When there’s any data for decision support or providing context to AI engines, all that comes under my remit. And for AI, it’s really, what do we buy, what do we build, who do we partner with. So there’s AI out of the box, and there’s also curated AI with the right data and context underneath.

WSJLI: Broadly, how does Virgin Atlantic think about using AI and agents?

Masters: We have 44 planes, a big brand, but we’re relatively small in terms of footprint as an airline. There’s so much we want to do and are doing to improve customer experience and premium experience for people. AI is a great way to help us do more and scale beyond our means.

With our data, it’s helping lots of other people across the organization access decision support that you would have traditionally got from a pricing optimization engine or machine learning.

The way I think about agents is, they’re a language model with additional context and access to tools. They’re experts that can then augment what people do.

WSJLI: How do you help employees choose between all the AI tools out there?

Masters: We’re pretty good at showing people the capabilities of the platforms I and the team have chosen, whether it’s Databricks or it’s OpenAI. If someone comes to us and says, ‘I want to use Claude,’ or ‘I want to use Dataiku or Snowflake,’ it’s like ‘Sure, but the capabilities are very similar here.’ The benefit of that relationship is beyond the pure tooling itself, it’s getting access early, it’s some of the co-marketing we’ve done.

We have a lot of the training pivoted around what problems you need to solve, ‘This is how you use the set of tools we’ve got to solve that problem,’ versus, ‘This is just what Databricks does.’

AI is freeing up hours at work. Now what? The percentage of frontline workers who say they use AI regularly jumped to 74% in 2026, up from 51% the year before, with many reporting that they are saving at least 8 hours per week, according to a Boston Consulting Group report. But what to do with that time is another matter: 66% report receiving limited or no guidance on the time they save, a gap that points to a leadership challenge.

Those findings come from BCG's AI at Work survey, which polled 11,749 workers across 14 markets and a broad range of industries. The WSJ Leadership Institute spoke with David Martin, an author of the report and the global head of BCG’s People & Organization practice, about some of the key takeaways. 

Don’t expect a four-day work week. Amid the uptick in AI adoption, 42% of frontline workers reported saving at least a workday per week. At the same time, 66% said their organizations give limited or no guidance on what to do with the time saved.

Satisfied, but strained. The survey also found improvements in job satisfaction among regular AI users: 57% of frontline employees, 67% of managers and 77% of leaders reported higher satisfaction.

But it may get better, he added, as companies think more about end-to-end workflows, creating "a more seamless employee experience."

The training gap has barely budged. Employees increasingly are incorporating AI into their work, but it’s not enough. Eighty-eight percent of employees believe they need major upskilling, according to the survey, yet only 36% feel properly trained.

"I think 2025 was '’let's invest in one hour, one day bootcamp,’  stuff like that, which just didn't move the needle clearly," said Martin "But I think companies have now realized that and you'll see that close hopefully."

— Tom Loftus

• Since launching its much anticipated AI model in April, Meta has delayed releasing an API — first from April to May due to bugs and infrastructure gaps, according to people familiar with the matter, then to June. Companies typically release an API alongside a new model, or within a few weeks of releasing it, to maximize its relevance among developers, the WSJ reports. 

• The CEOs of OpenAI, Anthropic, and Google DeepMind are among those urging Congress to require synthetic DNA and RNA suppliers to screen orders for dangerous combinations and verify customer identities, the WSJ reports. While the concerns are longstanding in the biotech industry, AI is magnifying them by potentially giving criminals the tools to unleash new pathogens.

• Broadcom logged strong quarterly revenue—fueled by demand for its AI chips—but left its 2027 outlook unchanged, sticking with guidance for AI semiconductor revenue of more than $100 billion. Shares in the semiconductor and software maker were down 13% in premarket trading Thursday.

• SpaceX bucked tradition Wednesday when it set a price for its shares over a week before its planned listing rather than setting a range and refining it after days of meetings with investors. The company said in a filing it plans to sell 555,555,555 shares at $135 apiece, raising close to $75 billion. It is targeting a valuation of around $1.75 trillion, The Wall Street Journal has reported.

• Benchmark, a Silicon Valley VC firm known best for early-stage bets on Uber and Twitter, plans to expand into investments in more mature startups, the WSJ reports. The firm's new funds include a $750 million fund dedicated to early-stage investments and a $1.25 billion fund for late-stage bets.