# Muse Went From Memo to Prototype in Two Weeks _Alexandr Wang told David Senra Muse went from June 2025 memo to February prototype to September launch after a seven-month reliability grind._ **Published:** 2026-10-11 **Source:** https://www.startuphub.ai/ai-news/artificial-intelligence/2026/muse-went-from-memo-to-prototype-in-two-weeks --- [Muse](https://www.youtube.com/watch?v=S6l3aRsecuE) began as a memo and turned into a working prototype in just two weeks, Alexandr Wang told David Senra, squeezing a vision for personal superintelligence into a seven-month push to ship something real. The idea predates the product by nearly a year. Wang said Meta Superintelligence Labs kicked off with a June 2025 memo titled Personal Super Intelligence, written alongside Mark Zuckerberg, Nat Friedman and a few others, that sketched out what Muse could become before agents were really a thing. Meta announced Meta Superintelligence Labs in 2025 pursuing 'personal superintelligence for everyone,' part of an industry-wide 2025-26 branding of superintelligence teams including Microsoft. What sealed the bet was February. Wang pointed out that Opus 4.5 had shown what agents could do, while OpenClaw was unfolding at the same time, and Friedman was the first inside MSL to go all in on OpenClaw. He told Wang the ride felt somewhere between terrifying and euphoric, watching him through security footage to nudging him to drink more water after he asked the agent to, a story Friedman later shared at Stripe Sessions. Wang said he tried Friedman’s deep psychoanalysis prompt himself after installing OpenClaw, with email and photo access feeding round after round of probing that seemed like three years of therapy packed into one session. Zuckerberg, according to Wang, felt the same mix of excitement and unease, using an agent at home with his kids and to review his own MMA training video for tips, while staff meetings that month turned into a show-and-tell of wild agent experiments. That moment set the timeline fork. Wang said the conventional wisdom in February was that [Anthropic](https://www.startuphub.ai/startups/anthropic) and Claude were far ahead on coding, and that coding agents were seen as the only worthwhile form factor. Meta took the opposite path, betting on personal agents because it already had the consumer reach and distribution to make them work. Wang noted that Meta touches three and a half billion people each day across Instagram, WhatsApp and Facebook, products built around what people want to do, not what they have to do, and that the promise of AGI is to shift time from the latter to the former. He argued Meta is uniquely positioned to build personal superintelligence for that reason, a view that matched the broader 2025 push where Meta, Microsoft and others all launched new teams branded around superintelligence. The next seven months were spent sanding down the edges. Wang said MSL whipped up a prototype in a week or two, showed it to the board, with Jollybot and an early version of Muse Charm already in the mix, then spent the rest of the time until September turning a fragile demo into a polished consumer product. The team logged more than 100 specific behaviors the model needed for a personal agent, built an evaluation for each, and ran review cycles to clear the red rows. Launch required a green light across the board, Wang said, and the first milestone that cleared it was a specially tuned version of Muse Spark 1.3. He said the product hinges on model quality, on how reliably it carries out tasks, communicates and checks back for guidance, and that OpenClaw’s trajectory, early magic then fading, revealed what happens when reliability never shows up, though it remains unclear if that consistency will hold when millions rely on it. The question that matters now isn’t the memo or the first prototype but whether that reliability can hold at scale. Wang credited Zuckerberg for staying patient through Wall Street skepticism from February to September, when the narrative was that Meta was burning cash and couldn’t win in AI, and said the work ahead is the same unglamorous grind, tweaking those measured behaviors until a nondeterministic system feels deterministic to billions. If Muse ends up the fastest-growing AI product since [ChatGPT](https://www.startuphub.ai/ai-news/market-research/2026/torsten-slok-apollo-ai-wage-study-pay-not-jobs), Wang says the reason isn’t a breakthrough but Meta’s refusal to ship until the spreadsheet turned completely green. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.