Alexandr Wang Bet Meta on Perception, and Muse Is the Test

What Wang told the founders

An interviewer asked whether Scale preferred to build its product or to sell it¹. Wang answered with Scale's own buyers in mind, a short list of very large customers, government agencies among them¹. "The reality is just so ugly most of the time that very rarely do people actually confront reality, and most of the time they just sort of choose to believe the perceptions," Wang said¹. From that he drew a job description for anyone selling to enterprises. "Just as much as your job is to improve the reality it is to shape the perception," he told the room¹.

Palantir, in Wang's telling, does that second job best. He called the company "a combination of a sort of acting troupe combined with a software company" and said it handed new hires an acting book, for a long stretch at least¹. A former Palantir engineer has since named the book. "I'm not sure if they still do this, but at the time when you joined they sent you a copy of Impro, The Looming Tower, Interviewing Users, and Getting Things Done," wrote Nabeel Qureshi, who spent eight years at Palantir¹⁷.

In Palantir's August shareholder letter, its chief executive described the sales force. "It must be noted that we have achieved these results with a minuscule and shrinking sales head count," Alex Karp wrote¹⁸. So the company Wang held up as the master of perception tells its own investors that its sales force keeps shrinking while its results grow.

The clip on X stops at the Palantir story. About 20 seconds later in the full recording, a voice on stage adds that a company still has to build a good product, and calls the perception work an extra layer on top of it¹. The automatic captions leave that speaker unlabeled, so the qualifier stays unattributed here.

Fifteen months inside Meta

Meta's deal, announced June 13, 2025, valued Scale at $29 billion, and Wang kept a seat on Scale's board as he left for Meta⁵. OpenAI dropped Scale as a data supplier within a week, saying it had been "seeking other providers for more specialized data to develop increasingly advanced AI models"⁷. On June 30, Mark Zuckerberg placed every AI team at Meta inside a new group, Meta Superintelligence Labs, with Wang as chief AI officer⁶.

Wang inherited a lab whose last model release had already cost it the trust of Meta's own leadership. Yann LeCun, Meta's chief AI scientist until he left to start his own company, told the Financial Times that Meta's researchers "fudged a little bit" when they tested the Llama 4 models against benchmarks⁹. According to Fast Company, the episode cost Meta's leadership, Zuckerberg included, its confidence in the team. LeCun judged his new boss bluntly: "There's no experience with research or how you practice research"⁹.

In October 2025, Meta cut about 600 jobs from the lab, mostly in its FAIR research unit and in infrastructure and product teams⁸. TBD Lab, the small group building Meta's next foundation models, kept its people⁸. Five months later, Meta pushed its next model, Avocado, from mid-March to May at the earliest. Internal tests showed it trailing Google's Gemini 3.0 and models from OpenAI and Anthropic in reasoning, programming and writing¹⁰.

Products arrived through the summer, and Zuckerberg told investors on July 29 that the lab's "trajectory is strong" a little more than a year after its launch¹¹. "Since we rebuilt Meta AI and integrated Muse Spark, we've seen a 60% increase in the number of people interacting with the assistant each day," Zuckerberg told analysts on the same call¹¹. Meta guided its 2026 capital spending to between $130 billion and $145 billion on that call¹¹. Scale, for its part, named Francis deSouza, a former Google Cloud executive, as its chief executive starting Aug. 10, 2026¹².

Muse and the market

Meta launched Muse in the United States on Sept. 8, and it became the top free download on both Apple's App Store and Google Play¹³. Meta's shares climbed from $613.48 on launch day to $777.59 on Sept. 24, a 26.8% gain in 16 days¹³.

Reuters found people doing part of the agent's work, in a test Meta ran on its own staff. Meta had tested a "human concierge" in which trained contractors placed some of the phone calls users asked Muse to make, Katie Paul reported Sept. 22¹⁴. About half of Meta's employees had the test switched on, and human callers succeeded on 95% to 98% of calls, above the rate for AI-only calls¹⁴. A vice president in Wang's lab called it "a miss" and said Meta had rolled the feature back¹⁴. "We're working with merchants to continue improving this potential calling feature, and will only roll it out when it's ready and with the proper disclosures," Meta spokesperson Daniel Roberts said¹⁴. Meta has run this experiment before: people handled an estimated 70% of tasks for M, its earlier assistant¹⁴.

The case for perception

Oren John, who makes ads at the creative agency Darkroom, tried Muse alongside OpenAI's Dots. "I tried Muse, I think Dots is better, and it doesn't really matter because the rollout of Muse was legendary and the rollout of Dots was the third of the big bots to come out so many weeks," John wrote on Oct. 1¹⁶. His advice to any rival follows from it: "You want to compete with any of these new tools, you have to be doing influencer marketing, and influencer at scale"¹⁶. John discloses that he has made content for AI companies, Meta AI among them¹⁶.

Justin Joseph, who teaches public relations at Boston University, sees a reason Meta gave Muse a cartoon mascot named Jolly. "I think a character lowers the psychological barriers to engaging with a new technology," Joseph told Fortune¹⁵.

Corporate budgets, measured across hundreds of companies, point the same way. MIT's study of enterprise AI found that more than half of generative AI budgets go to sales and marketing tools, even though the biggest returns showed up in back-office automation²⁰. By MIT's count, companies put most of their AI money where Wang put his emphasis, on the work of shaping how customers see them, ahead of the back-office work that paid best.

The case against

Ed Zitron, who founded the PR firm EZPR and criticizes the AI industry, reads the mascot as press bait. "It exists to get quoted in articles that say that 'Meta has given AI a cute face,'" Zitron told Fortune¹⁵. His test for Muse is a ranking, taken later: "Coming back in six months and seeing where Muse is on the iOS charts"¹⁵.

Aditya Challapally, lead author of the MIT study, located the companies that win in their execution. "It's because they pick one pain point, execute well, and partner smartly with companies who use their tools," he told Fortune²⁰. Menlo Ventures, which surveyed about 500 enterprise decision-makers, found that 47% of AI deals reach production, against 25% for traditional software¹⁹. Production is a test that happens on the buyer's own systems, after the launch campaign has ended.

What each side would need to be true

Wang's claim holds if Muse keeps its downloads and Meta keeps its share-price gain after users have run Muse and Dots side by side for months. It also holds if enterprise buyers keep spending on the agents with the loudest launches, whatever the production numbers say later.

Zitron's case holds if Muse slides down the iOS charts by spring, once the launch campaign ends. Karp's letter and Menlo's production rate offer an early sign for his side, since both describe buyers who judged software by what it did on their own systems¹⁸¹⁹. Reuters' account of human callers matters to both sides: the test stayed inside Meta, and Meta pulled it before any customer could rely on it¹⁴.

By the numbers

• 260,505: views on the Wang clip by Oct. 2, two days after it was posted⁴.
• 658 days: the gap between Wang's talk, Dec. 11, 2024, and the clip's spread on Sept. 30, 2026³⁴.
• $14.3 billion: what Meta paid for 49% of Scale AI in June 2025⁵.
• About 600: jobs Meta cut from its AI lab in October 2025⁸.
• 26.8%: Meta's share gain in the 16 days after Muse launched¹³.
• Up to 98%: the success rate for contractor-placed calls in Meta's internal Muse test¹⁴.

What to watch

Zitron's six-month test falls in March 2027, when the iOS charts will show whether Muse kept its users after the launch campaign¹⁵. Meta's next earnings calls will show whether the 60% rise in daily assistant users holds, and whether capital spending stays inside the $130 billion to $145 billion range¹¹. A public release of the Muse calling feature would test the disclosures Roberts promised¹⁴. Avocado, delayed in March, remains the clearest measure of Wang's lab on the frontier-model side, where benchmarks run in public and a launch campaign carries less weight¹⁰.

Sources

1. South Park Commons, "Scale AI & the Startup Hunger Games with Alexandr Wang," South Park Commons on YouTube, recorded Dec. 11, 2024, https://www.youtube.com/watch?v=CEebe451GZs
2. South Park Commons, "Alexandr Wang, Scale AI, & the Startup Hunger Games," South Park Commons, Dec. 13, 2024, https://blog.southparkcommons.com/p/alexandr-wang-scale-ai-and-the-startup-45f
3. South Park Commons, "-1 To Scale with Alexandr Wang," Luma, Dec. 11, 2024, https://luma.com/spc-alexandr-wang
4. Big Brain Business, "Meta Chief AI Officer Alexandr Wang on why your product can be better and still lose," X, Sept. 30, 2026, https://x.com/BigBrainBizness/status/2105311981783961941
5. Ivan Mehta, "Scale AI confirms 'significant' investment from Meta, says CEO Alexandr Wang is leaving," TechCrunch, June 13, 2025, https://techcrunch.com/2025/06/13/scale-ai-confirms-significant-investment-from-meta-says-ceo-alexandr-wang-is-leaving/
6. Rebecca Bellan, "Meta restructures its AI unit under 'Superintelligence Labs'," TechCrunch, June 30, 2025, https://www.techcrunch.com/2025/06/30/meta-restructures-its-ai-unit-under-superintelligence-labs
7. Maxwell Zeff, "OpenAI drops Scale AI as a data provider following Meta deal," TechCrunch, June 18, 2025, https://techcrunch.com/2025/06/18/openai-drops-scale-ai-as-a-data-provider-following-meta-deal
8. Adrita Shahriyar, "Meta lays off 600 jobs in its AI division," The Ticker, Nov. 3, 2025, https://theticker.org/17344/business/meta-lays-off-600-jobs-in-its-ai-division/
9. Mark Sullivan, "Yann LeCun: Meta 'fudged a little bit' when benchmark-testing Llama 4 model," Fast Company, Jan. 6, 2026, https://finance.yahoo.com/news/yann-lecun-meta-fudged-little-100000402.html
10. Matthias Bastian, "Meta delays its next AI model Avocado after internal tests show it can't keep up with Google and OpenAI," The Decoder, March 13, 2026, https://the-decoder.com/meta-delays-its-next-ai-model-avocado-after-internal-tests-show-it-cant-keep-up-with-google-and-openai/
11. Meta Platforms, "Q2 2026 Earnings Call Transcript," Meta, July 29, 2026, https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf
12. Scale AI, "Scale AI Appoints Francis deSouza as CEO," Scale AI, July 30, 2026, https://scale.com/blog/scale-appoints-new-ceo
13. Yoon Kyung-hwan, "Meta's 'Muse' AI Agent Tops U.S. App Charts, Lifts Shares 27%," Seoul Economic Daily, Sept. 26, 2026, https://en.sedaily.com/international/2026/09/26/metas-muse-ai-agent-upends-industry-in-us-debut
14. Katie Paul, "Exclusive: Meta testing a 'human concierge' for its new personal AI agent Muse," Reuters, Sept. 22, 2026, https://lufkindailynews.com/news_reuters/business/exclusive-meta-testing-a-human-concierge-for-its-new-personal-ai-agent-muse/article_2b93ac5a-2465-5821-8df6-47595cf3e8c3.html
15. Tatiana Sataua, "A Clippy marketing moment with Muse: 'Meta has given AI a cute face'," Fortune, Oct. 2, 2026, https://fortune.com/2026/10/02/meta-openai-ai-agent-mascots-jolly-dots-trust/
16. Oren John, "Silicon Valley's Marketing Arms Race," HYPER, Oct. 1, 2026, https://www.hyperstudios.us/p/silicon-valleys-marketing-arms-race
17. Nabeel S. Qureshi, "Reflections on Palantir," Oct. 15, 2024, https://nabeelqu.substack.com/p/reflections-on-palantir
18. Georgia Hennessy, "Alex Karp rings the alarm for salespeople, saying Palantir's teams crushed earnings with a 'miniscule and shrinking' head count," Business Insider, Aug. 4, 2026, https://finance.yahoo.com/technology/ai/articles/alex-karp-rings-alarm-salespeople-112230580.html
19. Menlo Ventures, "2025: The State of Generative AI in the Enterprise," Menlo Ventures, Dec. 9, 2025, https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
20. Sheryl Estrada, "MIT report: 95% of generative AI pilots at companies are failing," Fortune, Aug. 18, 2025, https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/