What is happening inside Meta's AI unit?

Last updated: 21 October 2026
Cover of NewMarketPitch's Frontier AI Labs Reality Check report, 2026 edition What is Meta spending on its AI unit, and what has it shipped? About $140B of spending planned in 2026 for all of Meta. Its new lab has shipped four Muse products, used by just over 3 million people a week. Chart from NewMarketPitch's Frontier AI Labs report.Do frontier AI labs with more money make better models? Only up to a point. The two richest labs make the best models, but several Chinese labs come close with a small fraction of the money. Chart from NewMarketPitch's Frontier AI Labs report.ChatGPT, Gemini, Meta AI, Claude: who has the most users? ChatGPT, with 1.2 billion people a week. Gemini and Meta AI each claim about a billion a month, and Anthropic gives no count for Claude. Chart from NewMarketPitch's Frontier AI Labs report.Who has held the top AI model, month by month? Anthropic and Google, taking turns: over the last 12 months Anthropic topped the LMArena users' vote for 7 months and Google for 5. OpenAI never did. Chart from NewMarketPitch's Frontier AI Labs report.Who is fighting whom in frontier AI? OpenAI leads Google on users, Anthropic leads OpenAI on sales, and Alibaba leads DeepSeek on downloads. Alibaba and Kling split video, SSI and Thinking Machines split money and product. Chart from NewMarketPitch's Frontier AI Labs report.Who is fighting on which front of frontier AI? Chat assistants are the most crowded front, with big tech, the US labs and three Chinese rivals. Europe shows up mostly through Mistral, on three fronts. Chart from NewMarketPitch's Frontier AI Labs report.How much have labs founded by OpenAI alumni raised? More than $130B, mostly by Anthropic, valued at $965B at its last private round. Labs founded by former Google researchers raised about $18B. Chart from NewMarketPitch's Frontier AI Labs report.Which lab makes the best open-weight models? Moonshot AI on test scores, with Kimi at 46 against 32 for DeepSeek and Zhipu. Alibaba's Qwen is downloaded the most. Chart from NewMarketPitch's Frontier AI Labs report.

Every funding round, every investor, and what the labs promised in return, in one document. Have a look ↓

Summary

Meta is trying to become an AI-first company faster than its own organization can comfortably absorb. Since June 2026, it has also started to show results.

The clearest change is on the model side. On October 12, 2026, Meta released Muse Wonder and Muse Wonder Full. On October 20, Arena ranked Muse Wonder Max second overall for text, and Artificial Analysis put Muse Wonder Full at 56, two points behind Anthropic’s leader. These are third-party estimates, but they are the first independent evidence that Meta is back near the frontier. Third-party rankings putting Meta near the frontier are why we separate what works today from what is still hype in our frontier AI labs reality check.

Meta also returned to open weights. On October 15, it released Muse Spark Open, an 816-billion-parameter model. It is the strongest US open-weight model by Meta’s own account, but outside indexes still rank it behind the best Chinese open models.

The organization underneath is still layered. Meta Superintelligence Labs sits at the top, with TBD Lab, FAIR, product teams, infrastructure teams, and a large Applied AI support unit of about 6,500 people, as WIRED reported in June 2026.

The data strategy still runs through several channels at once. Meta paid about $14.3 billion for its Scale AI stake in June 2025, but it also used rival vendors and now uses its own engineers and app users as data sources.

The employee backlash peaked in June 2026. WIRED and TechCrunch reported forced transfers, puzzle and coding tasks for model training, a “gulag” quote, and a petition against tracking. By September, Meta was reportedly inviting some Applied AI staff back into manager roles.

The spending level still explains the pressure. Meta now guides to $130 billion to $145 billion of 2026 capex. Q2 capex alone was $31.08 billion, and Q2 net income fell 14% even as revenue grew 28%.

Distribution is turning into a real product. The standalone Muse app launched on September 8, 2026 and reached about 10.9 million downloads by October 20, according to Sensor Tower estimates. Daily use is much smaller, at about 2.5 million.

So the story has moved. In June, the question was whether the reorg would produce anything. In October, Meta has a frontier-class model and a fast-growing app. The open question now is whether that output can justify the cost, human and financial, of getting there.

Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us

What is Meta spending on its AI unit, and what has it shipped? About $140B of spending planned in 2026 for all of Meta. Its new lab has shipped four Muse products, used by just over 3 million people a week. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is just a preview page. Don’t worry if a number looks old: the report gets refreshed, and you’ll always get the latest version.

How many AI teams does Meta actually have now?

Meta’s AI setup is not one clean lab. It is a layered organization, with a frontier-model group at the top and several support, research, product, and infrastructure teams underneath.

The main umbrella is Meta Superintelligence Labs, or MSL, led by Alexandr Wang. Inside it, TBD Lab focuses on frontier-model training, FAIR remains the long-term research group, Products & Applied Research turns models into Meta products, and MSL Infra handles the compute and systems layer. Meta's AI org chart keeps moving, and we go through what has changed over the past year in our report on frontier AI labs.

Then there is Applied AI, formed in March 2026 to support MSL’s researchers. In June 2026, WIRED reported that it had about 6,500 engineers and product managers. Other outlets put the number of employees moved into new AI units in May 2026 closer to 7,000.

That is a large share of the company. Meta had 75,472 employees at the end of June 2026, after a May reduction that affected about 8,000 people. On those numbers, Applied AI alone would be close to 9% of the workforce.

The structure also now has visible outputs. MSL shipped Muse Spark in April 2026, Muse Wonder on October 12, and Muse Spark Open on October 15. The consumer side got its own Muse app on September 8.

So the shape has not changed much since June. The difference is that it now ships frontier-class models: a small elite model team at the top, and a large human system underneath producing tasks, tests, and examples for its models.

Do frontier AI labs with more money make better models? Only up to a point. The two richest labs make the best models, but several Chinese labs come close with a small fraction of the money. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Just a preview. If a number here looks old, no worries: your report comes with the latest data.

What has Alexandr Wang delivered at Meta so far?

More than his critics expected in June 2026. In about 16 months, Wang’s organization has gone from reorg memos to a model that independent leaderboards place near the top.

The background is simple. In June 2025, Meta paid about $14.3 billion for a stake in Scale AI, valuing it at more than $29 billion. Wang left Scale to join Meta, stayed on Scale’s board, and became Chief AI Officer and head of MSL.

His reorg split the superintelligence effort into research, training, product, and infrastructure. TBD Lab sits closest to the frontier race. FAIR is more directly tied to model work. Products and Applied Research brings models into Meta’s apps. Infra exists because the whole strategy depends on huge GPU capacity.

The October 2026 results are the first real scorecard. Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index on October 20, up from 48 for Muse Spark, according to third-party coverage. That puts it above GPT-6 Astra and Gemini 4 Argon (53 each) and just behind Claude Opus 5.5 (58). On Arena’s overall text board, Muse Wonder Max ranked second, behind a preliminary Gemini 4 Argon result, with overlapping error margins.

Meta also confirmed on October 19 that the first phase of its Prometheus supercluster in Ohio is running, with hundreds of thousands of GPUs. It trained the Muse Wonder models there and plans to grow it toward about 1 million GPUs.

So Wang did not just add a senior title. He moved Meta from an open, research-heavy culture toward a tighter, centralized model-building culture, and that culture has now shipped a competitive model.

Did Meta basically buy Scale to fix its data problem?

Meta clearly wanted Scale’s data expertise, but the deal did not solve the whole data problem by itself.

The official transaction already says a lot. Scale said Meta’s investment valued the company above $29 billion and would “substantially expand” the commercial relationship around data solutions. The Financial Times put Meta’s outlay at about $14.3 billion.

But the follow-up signal is more revealing. In August 2025, TechCrunch reported that Meta’s TBD Lab was also relying on Scale competitors, including Mercor and Surge, for next-generation training work. Some researchers reportedly preferred those vendors’ data. Meta disputed that Scale quality was the issue, so that claim should not be overstated.

Still, the pattern is meaningful. If a company spends about $14 billion on a strategic Scale relationship and still works with rival vendors soon after, the bottleneck is not data in general. It is the exact kind of high-quality data needed for a specific model run.

In frontier AI, “data” no longer means a huge pile of internet text. It means hard coding tasks, reasoning problems, expert feedback, agent traces, safety tests, and edge cases that expose model weaknesses.

Meta now pulls that supply from several directions: Scale, rival vendors, its own employees, and, since September 2026, the Muse app and Muse Code, its paid coding product. We found no public report in October 2026 that changes the Scale picture, so the vendor story remains where it was in 2025.

ChatGPT, Gemini, Meta AI, Claude: who has the most users? ChatGPT, with 1.2 billion people a week. Gemini and Meta AI each claim about a billion a month, and Anthropic gives no count for Claude. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is a preview page. Spotted an outdated number? That’s normal here. The report you get is always the freshest version.

Are Meta engineers really being asked to do data-labeling work?

Yes, at least as reported in June 2026, but “data labeling” is not quite the right phrase.

According to WIRED, the work is closer to expert task-writing and model evaluation. Some Applied AI employees were reportedly asked to write puzzles that test how reliably models solve problems. Expert task-writing used to train models is a technology gaining traction that we follow in our frontier AI labs reality check.

Others were asked to produce complex coding problems that help AI scientists train and evaluate frontier models. Some employees were expected to complete two tasks per week.

That sounds modest until we look at who is doing it. These are engineers and product managers who may have been building Instagram, infrastructure, ranking systems, ads, messaging, or developer tools.

Now some of them create training and evaluation material for AI models. The work is more skilled than old-school click-labeling, but it can still feel like a demotion if your previous job was building products at global scale.

Another signal points the same way. Meta’s Model Capability Initiative, reported by Reuters and WIRED in spring 2026, records screen activity, clicks, keystrokes, and computer behavior in certain work contexts to train AI agents. Employees pushed back hard, and WIRED reported that an internal protest post was seen by nearly 20,000 coworkers.

We found no October 2026 report that these task programs have ended. The direction set in June still holds: Meta is turning employee knowledge and employee behavior into AI training material.

Why would Meta put strong engineers on puzzle tasks?

Because the easy data is no longer enough.

Older AI systems improved mostly with more general text, labels, and human preference data. Today, the frontier fight is more specific. Models need harder reasoning tests, harder coding tasks, better agent traces, richer safety evaluations, and examples of how humans actually complete work on a computer.

The October 2026 models show where that data goes. Meta pitched Muse Wonder on coding, planning, research, and long agent tasks, with features it calls Active Memory and Active Focus. It promoted Muse Spark Open on tool-calling benchmarks. These are exactly the areas where generic data helps least.

The weak spots point the same way. On Terminal-Bench 2.1, a pure coding test, Kimi K3 scored 85 versus 76.4 for Muse Spark Open, according to TechRepublic’s account. If a company wants to close that kind of gap, it needs hard coding tasks written by people who know how to code.

That is why the work can be strategically useful and frustrating at the same time. From the model team’s side, a good engineer writing two hard coding problems per week is valuable. From the engineer’s side, it can feel strange: you joined Meta to build products at scale, and now your work becomes input for a model team.

Who has held the top AI model, month by month? Anthropic and Google, taking turns: over the last 12 months Anthropic topped the LMArena users' vote for 7 months and Google for 5. OpenAI never did. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Preview only. Some numbers may have changed since we made this page, but your report will be up to date.

Is the “AI gulag” line real or just drama?

The phrase is dramatic. The frustration behind it, as reported in June 2026, looked real.

In June 2026, WIRED quoted one Applied AI employee calling the unit “literally the gulag,” and TechCrunch summed it up as a “soul-crushing gulag.” That is an anonymous employee using extreme language, not a literal description of working conditions.

But the surrounding facts explained the anger. The unit was new, had about 6,500 people, and many were reportedly moved in by surprise email, with little real choice except to join or leave.

Several morale signals pointed the same way that month. WIRED reported a livestream incident where an employee interrupted an internal presentation. It reported that some Applied AI workers called themselves “draftees.” More than 1,600 employees reportedly signed a petition against the employee-tracking initiative, and a protest post about it was seen by nearly 20,000 employees.

What happened next suggests Meta heard it. In May 2026, Zuckerberg wrote that he did not expect other company-wide layoffs that year, and in June he reportedly admitted mistakes in the reorgs. In September 2026, Business Insider and Fortune reported that Meta was inviting some Applied AI individual contributors to return to manager roles. That is a partial reversal of the flattening that fed the anger.

So the “gulag” line was the part that traveled online. The real story was forced redeployment, surveillance anxiety, layoffs, and very wide manager ratios. By October 2026, the loudest phase seems to have passed, but we found no sign that the underlying model of work has changed.

Is Meta back at the AI frontier with Muse Wonder?

Close to it, on independent measures. That is a real change from June 2026, when Meta’s best model was still clearly behind.

The timeline is fast. Muse Spark launched in April 2026 as MSL’s first model. Axios reported it was code-named Avocado and built over nine months, and Meta admitted gaps in areas like coding. Six months later, on October 12, Meta released Muse Wonder and Muse Wonder Full, which it calls its most capable models so far. Six months from admitted gaps to Muse Wonder is a fast turn, and we weigh who the likely future winners are in our report covering frontier AI labs.

Third-party scores back that up. On October 20, Arena ranked Muse Wonder Max second overall for text, at 1,510, between a preliminary Gemini 4 Argon result (1,524) and Claude Opus 5.5 (1,509). The error margins overlap, so this is a tie zone, not a clear ranking. Artificial Analysis put Muse Wonder Full at 56, behind only Opus 5.5 at 58.

The open-weight side is weaker. Muse Spark Open, released October 15, has 816 billion parameters, 42 billion of them active, and a 1 million-token context. Apple supports running it locally through MLX. But it scored 39 on the Artificial Analysis index, behind Kimi K3, GLM-5.3, and Xiaomi’s best open model, according to TechRepublic.

Usage is growing but still small. The Muse app launched on September 8 and hit No. 1 on the US App Store within ten days. Sensor Tower estimated 10.9 million downloads by October 20, 10.2 million of them in the US, and about 2.5 million daily users on October 16. Meta has not disclosed official usage figures.

So yes, Meta is back near the frontier on capability. It is not yet ahead, its open model trails Chinese rivals, and its app turns only a fraction of downloads into daily habits.

Who is fighting whom in frontier AI? OpenAI leads Google on users, Anthropic leads OpenAI on sales, and Alibaba leads DeepSeek on downloads. Alibaba and Kling split video, SSI and Thinking Machines split money and product. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is just a preview page. Don’t worry if a number looks old: the report gets refreshed, and you’ll always get the latest version.

Can Meta’s AI products justify $145 billion of capex?

Not yet, on any visible revenue line. Meta is buying time, compute, talent, and data at once, and paying for it with ads.

The numbers keep rising. Meta now guides to $130 billion to $145 billion of 2026 capex, up from $125 billion to $145 billion in April. Q2 capex was $31.08 billion, including finance lease payments. Some outlets cite about $30.1 billion, which excludes those payments. Either way, it is about 1.5 to 1.6 times the $19.84 billion spent in Q1.

That capex is companywide, not only AI research. But Prometheus shows where much of it goes: a cluster with hundreds of thousands of GPUs today and about 1 million planned.

The cost is now visible in profits. Q2 2026 revenue was $60.8 billion, up 28%. But net income fell 14% to $15.85 billion, and operating margin dropped to 31% from 43%. Part of that came from $2.40 billion of legal charges and $1.18 billion of severance, so not all of it is AI.

The return is still small next to the bill. Muse Wonder costs $4 per million input tokens and $20 per million output tokens on Meta’s API, the same list price as Claude Opus 5.5. Muse paid plans run $20 and $100 a month. Meta has not disclosed revenue or subscriber numbers for any of them.

Investors have so far accepted the trade. Meta shares jumped about 11% on September 22, two weeks after the Muse app launch, and rose about 3.4% on the day Muse Wonder launched.

So the spending is not just a financial detail. It explains why the internal transition feels so intense. When capex reaches this level and profits start to fall, every team gets reorganized around AI.

What is proven, and what is still only reported?

The cleanest way to read Meta AI is to separate official facts, independent estimates, strong reporting, and contested claims.

The official facts are solid. Wang is Chief AI Officer. Meta released Muse Spark in April, Muse Wonder on October 12, and Muse Spark Open on October 15. Prometheus is running. Q2 2026 revenue was $60.8 billion, capex $31.08 billion, headcount 75,472, and 2026 capex guidance is $130 billion to $145 billion. About 8,000 employees were affected by the May reduction.

Then there are independent estimates. Arena and Artificial Analysis rankings are third-party measurements with error margins. Muse app downloads and daily users come from Sensor Tower and are not company figures. Two download estimates differ slightly: about 10.4 million by October 16 in a BNP Paribas analysis, and about 10.9 million by October 20 in Sensor Tower data.

Then there is strong reporting. In June 2026, WIRED said Applied AI was formed in March, had about 6,500 engineers and product managers, and included staff writing puzzles and coding problems for model training. WIRED also reported the livestream incident, the “draftee” language, the 1,600-plus petition against tracking, and 50-to-one manager ratios in some teams. In September 2026, Business Insider reported the offer to return some staff to manager roles.

The more delicate part is still the vendor-quality story. TechCrunch reported in 2025 that some TBD Lab researchers preferred data from Surge and Mercor, while Meta pushed back. We can say there were signs of tension in Meta’s data supply chain. We should not say that Scale “failed” Meta.

Overall, the evidence base is stronger than in June. The models and spending are official, the model quality is independently measured, and the employee anger is strongly reported, but mostly as of June 2026.

Who is fighting on which front of frontier AI? Chat assistants are the most crowded front, with big tech, the US labs and three Chinese rivals. Europe shows up mostly through Mistral, on three fronts. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Just a preview. If a number here looks old, no worries: your report comes with the latest data.

Is Meta copying OpenAI, or doing something different?

Meta is borrowing the frontier-lab playbook, but it is applying it inside a very different company, and in October 2026 it added an open-weight track that OpenAI does not lead.

The OpenAI-style part is clear: a superintelligence lab, a small elite model team, huge compute, expensive hiring, reasoning models, and a paid coding product. Muse Code and the Muse Wonder API compete directly with Codex and Claude Code. TBD Lab is Meta’s inner model room. With Muse Code now facing Codex and Claude Code, we sort out who is ahead, and who is slipping, in our report on frontier AI labs.

But Meta has something most labs lack: distribution across WhatsApp, Instagram, Facebook, Messenger, Threads, smart glasses, and about 3.6 billion daily active people in June 2026. Meta used it. It gave Muse up to half of its daily house ad impressions in mid-September, and downloads jumped.

The open-weight bet is the other difference. Muse Spark Open runs on Hugging Face, AWS, Google Cloud, and Microsoft Foundry, and Meta pitches it as a US alternative for companies that cannot use Chinese models. Its license still requires a custom deal for large model-serving businesses, so it is open, but not without conditions.

The tension is unchanged. Meta is trying to operate like a frontier lab while still being a giant social and ads company. One culture rewards small elite teams and fast model runs. The other has thousands of engineers, privacy constraints, moderation issues, ads incentives, and consumer UX trade-offs.

Why are Meta employees angry while revenue keeps growing?

Because Meta can be commercially strong and culturally unstable at the same time. In 2026, the strength is in revenue, not in profit.

The top line is still excellent. Q2 2026 revenue grew 28% to $60.8 billion. About 3.6 billion people used Meta’s apps daily in June. But costs grew 55%, net income fell 14%, and operating margin fell to 31%. Meta is healthy, but it is now paying for AI out of profits that employees can see shrinking.

Inside the company, the experience has been harsh. About 8,000 people were affected by the May 2026 reduction. About 6,500 to 7,000 were moved into new AI units, depending on the report. Employees dealt with tracking software, 50-to-one manager ratios in some teams, and teams being dissolved or merged. Business Insider reported that Meta had dissolved two major AI teams in less than six months.

Meta has softened some of it. Zuckerberg said he did not expect other company-wide layoffs in 2026 and reportedly admitted mistakes. In September, Meta reportedly began inviting some Applied AI staff back into manager roles. Those are signs the company knows it pushed too hard.

In the end, Meta’s AI push is funded by a strong ads business, but the organization has been reshaped as if the company were at war. The October model results give that war a first win, not an end.

How much have labs founded by OpenAI alumni raised? More than $130B, mostly by Anthropic, valued at $965B at its last private round. Labs founded by former Google researchers raised about $18B. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is a preview page. Spotted an outdated number? That’s normal here. The report you get is always the freshest version.

So what is really going on at Meta AI right now?

Meta is trying to become an AI-first company faster than its organization can comfortably absorb. In October 2026, it has the first proof that the approach can work.

The pieces now line up. Meta hired Wang from Scale and created MSL. It split the AI org into training, research, product, and infrastructure. It built a 6,500-person Applied AI unit and collected employee work data for AI agents, then faced a major backlash in June 2026. It narrowed 2026 capex guidance to $130 billion to $145 billion. It brought Prometheus online. And between September 8 and October 15, it launched the Muse app, Muse Wonder, and Muse Spark Open.

Taken together, this is not just “Meta doing AI.” It is a company-wide conversion, using money, people, compute, product distribution, internal work data, and vendor data to catch up in frontier AI.

The result is mixed but clearer than in June. On capability, Meta is now in the top group, close to Anthropic and Google on independent scores. On open weights, it is back, but behind the best Chinese models. On usage, the Muse app grew fast, but daily use is a fraction of downloads. On money, Meta has not shown AI revenue that matches its spending.

The human layer is still the uncomfortable part. Behind the leaderboard scores are engineers writing coding tasks, employees being tracked to train agents, and managers absorbing very large teams.

So today, the clearest answer is this: Meta AI is not empty and not failing. It is spending massively, shipping frontier-class models, and carrying real internal strain. The June story was the backlash. The October story is that the strain produced results. The next test is whether those results turn into revenue before the bill gets harder to defend.

Which lab makes the best open-weight models? Moonshot AI on test scores, with Kimi at 46 against 32 for DeepSeek and Zhipu. Alibaba's Qwen is downloaded the most. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Preview only. Some numbers may have changed since we made this page, but your report will be up to date.

Our methodology

We approached the Meta AI question as an aggregation problem rather than a single-news-event story. The public picture is confusing because Meta’s AI shift is showing up through several different signals at once: leadership changes, org design, data-supply moves, employee redeployment, product launches, infrastructure spending, and internal backlash.

To avoid a vibe-based answer, we broke the question into those analytical dimensions and looked for the freshest available signals in each one. We gave more weight to first-hand disclosures, official company materials, financial filings, and recent reporting from strong technology and business outlets. We also used independent model rankings (Arena, Artificial Analysis) and app-intelligence estimates (Sensor Tower), and we label them as estimates.

Where a signal was reported rather than official, we treated it as evidence of internal dynamics rather than as a settled company position. That distinction matters especially for employee morale, vendor-quality claims, and internal reorg details, much of which was reported in June 2026.

The final answer comes from the convergence across those dimensions. Meta is not just launching models, not just reorganizing teams, and not just spending heavily on infrastructure.

The clearer picture is that the company is trying to convert itself around frontier AI, using capital, talent, compute, internal labor, product distribution, and data supply chains at the same time. That structured aggregation is what makes the conclusion stronger than any one quote, reorg memo, product launch, or capex number on its own.

Key sources used for this analysis include: Meta’s Q1 2026 results, Meta’s Muse Spark announcement, Meta’s Muse Spark evaluation methodology, Meta’s Muse Spark safety and preparedness report, Meta’s Advanced AI Scaling Framework, Meta’s Alexandr Wang leadership page, Scale AI’s announcement on its Meta relationship, Axios on the Meta-Scale deal, Axios on Muse Spark, WIRED on Meta’s internal AI meeting and employee backlash, WIRED on employee protest over tracking for AI training, TechCrunch on Meta’s Scale partnership tensions, TechCrunch on Applied AI employee frustration, The Verge on Meta’s employee-tracking AI initiative, Ars Technica on employee tracking for AI agents, Business Insider on Muse Spark, Business Insider on layoffs, manager ratios, and AI spending, TechSpot on Meta’s AI restructuring, TechRepublic on Muse Spark Open, Verdict on Meta’s Q2 2026 results, The Next Web on early Muse usage, Fortune on Meta bringing back managers in Applied AI, TIKR on Muse Wonder and Meta’s capex, AI Weekly on the Muse Wonder release, Let’s Data Science on Muse app downloads, Outlook Business on Meta ruling out more company-wide layoffs, AI Weekly on the Applied AI transfers, AI Weekly on Applied AI staff returning to manager roles, and PR Newswire’s republication of Meta’s Q1 2026 results.