Source: raw/You_ve_Never_Lost_19_Billion._So_Why_Are_You_Playing_It_Safer_Than_Zuckerberg.md
Creator: Nate B Jones, AI News & Strategy Daily · URL: https://www.youtube.com/watch?v=JFoDdKuRmSk · Duration: not recorded · Platform: YouTube (auto-captions, fetched 2026-11-05).
A motivational episode with a few concrete product lessons. Jones points out that Meta’s Reality Labs lost $19.2 billion in 2025 alone after Zuckerberg renamed the company for a metaverse most people never joined, and that Meta then made another platform bet with Muse, which he says 10 million people set up in six weeks. He draws three things a builder can copy: who Meta put on it (a small team), the polish in parts of the product nobody demos, and trying again after a miss with something actually changed. His closing argument is that AI has lowered the cost of attempting things but not the embarrassment of failing, and that embarrassment is what keeps people’s ambitions small.
Key Takeaways
- Muse figures he gives. 10 million people set Muse up in six weeks. Per Alexander Wang (in an October interview with David Senra), fewer than 200 people worked across the model and the product. Wang described more than a hundred specific model behaviours the team tested and improved before launch. Jones: “Muse is not the smartest agent on the market. That’s not the point.” He also says Muse is expensive to serve and Meta must bring that cost down to reach a billion users.
- Keep the team small on purpose. Wang’s team stayed small by big-company standards so hard product decisions were made sharply rather than “resolved… by weight of consensus”. Users feel those compromises: the same thing behaving differently in two tabs, a second set of rules for a closely related task. Then let the people you picked “cook”.
- Polish where nobody looks. His test is a restaurant’s bathroom: anyone cares about the entrance; caring for the dull parts shows everything is cared for. An assistant you live with daily makes small rough edges irritating fast.
- Show the user what the agent is doing. His favourite detail in his own Muse (named Marvin): it gets out a little laptop and types, or scratches its head, so he never reads a technical status message. Whoever builds an agent faces the same question throughout: has the user understood, is the agent waiting on the user, is the task finished? “Every time a person has to stop and think and investigate… the magic disappears.”
- Character is not enough. Meta built a model for live avatar expressions synced to speech ahead of a coming voice experience and a wearable, but “a charming character that fails to do the job is still not good enough.”
- His own result. Marvin helped him cancel $1,285 a year in subscriptions he had been putting off.
- Change something before the next swing. Coming back with the same assumptions and more enthusiasm will not fix a failed project. Change who you work with, your approach, or how you understand the customer. He asks of his own disappointing projects: what would actually make my next attempt better?
- Stick with one idea through the dull work. Fast prototyping tempts you to keep making first passes; people only keep using something once you have fixed its awkward parts.
- Show instead of pitching. His example: someone who has spent years in an industry where customers cannot compare complicated quotes can now build a working comparison tool from real quotes and show it to people, instead of picturing a funded development team. Scaling may still need technical help.
- Embarrassment did not get cheaper. “The technical and financial barriers have come down, but the emotional barrier is exactly where it was.” Revisit the ideas you ruled out because the first version needed a team you could not afford. Pick work you would get more excited about when the next model ships, not more worried.
Try It
- Reopen one idea you shelved because the first version needed money or a team, and build the smallest version someone can use.
- If you build an agent, design its status first: how the user can tell at a glance that it is working, waiting on them, or done.
- Do a “bathroom pass” on your product: use the screens nobody demos and fix the inconsistencies.
- Before retrying a failed project, write down what you are changing (people, approach or customer understanding). If nothing, don’t retry yet.
Open Questions
- The 10 million figure is Jones’s; its source is not given. The fewer-than-200 and hundred-behaviour figures come from Wang via Jones (the captions spell the name “Wayne” once).
- Reality Labs’ 19 million” once, apparently a slip for billion.
Related
- Personal Agents Compared (October 2026) — Muse’s design notes, including its status-first decision
- Claire Vo’s Personal Agent Benchmark — Muse against Grok Bot, Dots and Duet on nine tasks
- OpenAI’s Vijay Raji — the same “excited, not worried, when a model ships” test
- AI-Proof Businesses and the Moat Question — the defensive version of the same question
- What Personal Agents Keep — the access a personal agent like Muse needs