Meta claims AI will accelerate app launches, but decades of failed experiments raise questions about whether building apps was ever the company's biggest hurdle.
Meta told investors this week that artificial intelligence has rewritten the cost of building software, and that a run of new consumer apps will follow. On the company's second-quarter 2026 earnings call Wednesday, chief executive Mark Zuckerberg said large language models now let his engineers ship products at a faster clip, and that Meta intends to lean on its recommendation systems to carry those products to the users most likely to stick with them.
"I expect it to become a lot easier to ship new apps," Zuckerberg said. He added that Meta is "planning to build out more ideas" and that unnamed "new consumer products" were "releasing soon."
The message landed inside an earnings report that Wall Street mostly read as a warning about spending rather than a promise about products. Meta's stock fell after hours even as revenue climbed, and analysts spent their questions on capital expenditure and enterprise strategy instead of the app pipeline. The apps, for now, are a subplot.
But it is a subplot with a long history at this company, and that history is the reason to read the claim carefully.
Zuckerberg tied his optimism to products Meta has already pushed out. Earlier this year the company shipped Instagram Instants. It followed with Forum, a standalone app for Facebook Groups, and Seller, a standalone app built for Marketplace vendors.
Around those came a gaming app assembled through so-called vibe coding and a separate photos app spun out of Instagram. Meta also ran an experiment that generates AI bedtime stories for children.
None of these is a household name yet. That is partly the point Zuckerberg is making: when the cost of a first version drops, a company can afford to throw more attempts at the wall and see which ones hold.
The theory rests on two pieces working together. Cheap building supplies the volume of experiments. The recommendation engine supplies the distribution that decides which experiment finds an audience. Meta has spent two decades tuning the second half of that machine, and it now argues the first half has caught up.
The app push sits alongside a faster cadence of model releases. In the past month Meta shipped Muse Image, an image model, plus an update called Muse Spark 1.1, both out of the Superintelligence Labs group Zuckerberg stood up a little over a year ago. The pitch is consistency: the same AI progress that writes better models is supposed to write more apps and rank them better once they exist.
Here is the part the earnings call skipped.
This is the third time Meta has bet on a portfolio of standalone apps, and the first two rounds ended in near-total shutdown.
In its Facebook years the company ran an internal incubator called Creative Labs, which pushed out a photo-sharing app called Slingshot, an anonymous chat app called Rooms, a news reader called Paper, a photo app called Moments, and a collaborative video tool called Riff. By 2015 the effort was wound down and the apps were retired one by one, unable to find crowds of their own.
The early 2020s brought a second attempt through a research group known as the NPE Team. Its output was longer and stranger. The list ran through a chat app called Bump, a social music app called Aux, a task manager called Move, a dating app called Spark, a calling app called CatchUp, a zine maker called E.gg, an events app called Venue, a creator Q&A app called Hotline, a Cameo-style app called Super, a couples app called Tuned, and a music app called BARS. Every one of them was eventually closed.
Two clean-outs, dozens of apps, and no breakout among them.
The engineering was never the thing that failed. Getting strangers to open an unfamiliar icon a second time was.
That is the weak point in the "AI makes building easier" framing. Building was rarely Meta's bottleneck. Distribution and retention were, and a language model does not fix those on its own.
Meta does have one recent win, and Zuckerberg reaches for it every time this subject comes up.
Threads, the company's answer to Elon Musk's X, crossed 500 million monthly active users in June and has held a lead over X in worldwide daily actives for months. Zuckerberg has said he expects it to become Meta's next platform with a billion users.
Threads did not win by being built quickly, though. It won because Meta seeded it directly from Instagram's logged-in base and then promoted it across Facebook and Instagram for two years straight. The growth story there is a distribution story wearing an app's clothing.
AI did play a role, just a different one from the "faster building" pitch. CFO Susan Li told investors that language models are increasingly driving gains in ranking and recommendations, the systems that decide what surfaces in a feed. She said every Reel and Feed post on Instagram now passes through an LLM that reads it for topic and tone before the recommendation system decides who sees it. Li added that the company sees room to keep improving those recommendations into 2027.
So there are two separate AI claims folded into one upbeat sentence on the call. One is that AI helps write the apps. The other is that AI helps distribute them. Threads is real evidence for the second. It is thin evidence for the first.
The app talk shared a stage with a spending story that dominated the reaction.
Meta reported second-quarter revenue of $60.8 billion, up 28% from a year earlier, beating estimates on the top line. Earnings told a rougher story. Per-share profit came in at $6.18, well short of the roughly $7.15 analysts expected, dragged down by a $2.4 billion legal charge and about $1.2 billion in severance tied to a May headcount cut of around 8,000 people.
The larger shock was cash. Capital spending hit $31.08 billion for the quarter, more than double the $17 billion Meta spent in the same period a year earlier, almost all of it aimed at data centers and AI infrastructure. Free cash flow collapsed to $784 million, down from $8.5 billion a year before. Operating margin slid to 31% from 43%.
Then management raised the full-year capital-expenditure range to between $130 billion and $145 billion, lifting the floor by $10 billion in a single quarter.
Shares fell somewhere between 7% and 10% in after-hours trading, near the bottom of the stock's 52-week range.
Meta is not the only company in this position. Alphabet and Amazon have both pushed capital-expenditure targets sharply higher on AI, and Meta's own quarterly figure now sits in the same neighborhood as Alphabet's. What unsettled investors was the combination: record revenue that still could not keep free cash flow from nearly vanishing.
Set against that backdrop, a handful of experimental consumer apps was never going to be the headline. Investors want to know when the AI bill starts paying for itself, and a bedtime-story generator is not the receipt they are looking for.
Analysts on the call barely touched the app pipeline. Their questions clustered on two things: how much Meta will spend on AI, and how it plans to earn that money back from something other than advertising.
That second question is where Zuckerberg spent his energy. He sketched an enterprise business built on APIs and agents that run inside WhatsApp and Messenger, with raw compute sold to outside customers further out. Meta said more than 1 million businesses were already using its business agents each week across its messaging apps. Family of Apps "other" revenue, the bucket that captures paid messaging and subscriptions, reached $1 billion in a quarter for the first time, up 73% from a year earlier.
Zuckerberg argued that selling intelligence built on top of compute should carry higher margins than renting out the compute itself, and said Meta has already fielded offers to buy capacity at a premium to what it paid. He declined to name which of these bets would scale first, saying only that he expected "meaningful growth in all of these areas."
Zuckerberg's forecast is specific enough to be tested. He said more consumer products are close, and that Meta will use its recommendation systems to scale the ones that show early signs of life, the same playbook that carried Threads.
The open question is whether any of the recent launches gets that treatment or quietly disappears the way the Creative Labs and NPE apps did. Cheap experiments are worth something only if the company commits its distribution muscle behind the ones that work, and most experiments will not earn it.
For a company spending north of $130 billion a year to build the future, a few experimental apps are a low-stakes side bet. For the strategy of "ship more, scale the winners" to mean anything, Meta has to produce a winner that was not hand-fed from Instagram's existing audience. It has not done that yet.
OpenAI pauses Astra after safety tests suggest the unreleased AI model may have reached a critical l...
SaferAI's report reveals China's GLM-5.2 rivals frontier AI models but lacks meaningful safety guard...
Apple signals that heavy Siri AI users may need iCloud+ to unlock higher usage limits. Here's what T...
Discussion
Join the discussion and share your perspective.