Zuckerberg bets the rest of Meta's decade on agents the company has not yet shipped

On Meta's second-quarter 2026 earnings call on Wednesday 29 July, chief executive Mark Zuckerberg told investors that the company is preparing a "big push" into personal artificial-intelligence agents that can act on a user's behalf, and that he expects "billions of people" to have such agents within five years. The Verge, reporting from the call, said Zuckerberg framed the agents as the next consumer platform; TechCrunch, also on the call, recorded the five-year horizon and the broader enterprise pitch; the BBC reported the same hour that Meta shares fell as analysts pressed management on the cost of the build-out and on what, exactly, the company intends to sell.

Monexus analysis: the call's most consequential sentence was not about agents at all. It was Zuckerberg's commitment, also reported by the BBC, that Meta intends for the first time to sell AI tools to other companies. That is a strategic reorientation, not a product update. A firm that built its margins on ads sold against user content is now publicly telling the market that the next leg of growth will run through enterprise contracts, third-party developers, and rented compute. The personal-agent story is the consumer wrapper; the enterprise story is the cash register.

The pitch, as it landed

Zuckerberg's argument, in the order he made it, ran like this. Meta's Q2 results justify continued heavy spending on AI infrastructure because the company's family of apps, Facebook, Instagram, WhatsApp and Messenger, gives it a uniquely large surface area on which to deploy agents at scale. The Verge's write-up of the call said the CEO positioned agents as software that will "do things on your behalf," a deliberately vague formulation that covers everything from booking restaurants to negotiating with merchants. TechCrunch's separate read of the same call added the explicit forecast: "billions of people will have personal AI agents in five years."

The pitch landed unevenly. The BBC's market report, filed the same evening, said Meta shares fell as "frustration grows over AI spending plans," and quoted Zuckerberg's confirmation that Meta will, for the first time, sell AI tools externally. The combination, an elevated capex line and a less familiar revenue line, is what usually moves the multiple. Investors were not, on this evidence, fully sold.

What "selling AI tools to other companies" actually means

Monexus assessment: read literally, the enterprise opportunity Zuckerberg described on the call is four distinct businesses stacked on top of each other, and Meta's competitive position in each is different.

The first layer is agents themselves, packaged products that other firms can license or fine-tune. The second is application programming interfaces, the connective tissue that lets outside software call Meta's models, and the third is compute, raw access to the GPUs Meta is buying in volume. The fourth, and least remarked in the wire coverage, is internal software: the tooling Meta has built for its own engineers, which it now wants to productise. TechCrunch reported Zuckerberg describing the bundle as a "large enterprise opportunity" spanning exactly those four pieces.

The competitive read is harder than the marketing. On agents, Meta is one of several incumbents, alongside OpenAI, Anthropic, Google and Microsoft, all chasing the same enterprise procurement budget. On compute, the company is competing with Amazon Web Services, Microsoft Azure and Google Cloud, all of whom sell to the same buyers and own more of the data centre stack. On APIs, Meta's open-weight Llama strategy has given it developer mind-share, but mind-share is not the same as revenue. Monexus finds that the enterprise story is plausible, but only if at least one of the four layers converts at scale, and the call did not give investors a basis to pick which one.

The capital question no one on the call could answer cleanly

Monexus analysis: underneath the agent narrative is a capital-allocation problem. The BBC's report made the source of investor frustration explicit: the spending plans are large, the revenue line is new, and the timing is the sell-side analyst's worst combination. Zuckerberg has separately, according to a post on X by the account Unusual Whales, "requested that the US must accelerate AI development," a framing that puts Meta's capex inside a national-competition story rather than a return-on-invested-capital story. Whether that framing helps or hurts the multiple depends on which investor is asked.

Two structural facts make the call harder to read than its peers at Alphabet or Microsoft. First, Meta is not a cloud business with an AI wrapper; it is an ads business with an AI problem, and the market values the former at a different multiple than the latter. Second, the company is funding the build-out out of operating cash flow rather than from a separate enterprise revenue line, which means every dollar of GPU spend is also a dollar not returned to shareholders. The earnings call did not surface a clean answer to either of those, and the share price reaction, as the BBC recorded it, suggests the market noticed.

What the next quarter will and will not tell us

The honest reading of the call is that Zuckerberg has chosen the strategy and named the timeline, and the only question left is whether Meta can ship. Five years is long enough for the agent market to be reshaped twice, and short enough that the capex bill will arrive before the agent revenue line has matured. The Q2 print did not break the thesis; it did, however, raise the bar for what Q3 and Q4 will need to show.

Three things to watch in the next two earnings cycles. First, any disclosure of agent-related revenue, even as a percentage of total, would be the cleanest signal that the enterprise story is converting. Second, capital-expenditure guidance, which the BBC's report flagged as the proximate cause of the share-price reaction, will be the lever that moves the stock most directly. Third, the developer signals around Llama and any successor models: download counts, fine-tune volumes, and named enterprise customers are the ground-truth indicators the call itself did not provide.

The sources cited above do not specify how Meta intends to price agents for consumers, what the unit economics look like, or which enterprise customers, if any, have already signed. Those gaps are not unusual for a strategy-stage call, but they are the gaps investors will want closed before the share price recovers.

Desk note: Monexus treated the Verge, TechCrunch and BBC reports of the same earnings call as three independent reads of the same event, and labelled the strategic-reorientation argument as analysis rather than reporting. The Unusual Whales post was treated as a low-weight framing signal, not as a primary source on policy.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://www.theverge.com/tech/972294/meta-q2-2026-earnings-mark-zuckerberg-personal-ai-agents
• https://www.bbc.co.uk/news/articles/ckgd31l5yrdo?at_medium=RSS&at_campaign=rss
• https://techcrunch.com/2026/07/29/mark-zuckerberg-predicts-that-billions-of-people-will-have-personal-ai-agents-in-five-years/
• https://techcrunch.com/2026/07/29/zuckerberg-says-metas-enterprise-ai-opportunity-extends-beyond-agents/
• https://t.me/theverge_news/8093
• https://x.com/unusual_whales/status/2082490581968977927