
Episode #22
Where Are the Humans?
Dear show notes readers, This week we accidentally spend most of an hour asking the same question in several completely different ways: Where are the humans? We start with personal AI agents and what happens when they begin doing more of our searching, shopping, comparing, emailing, and general internet busywork for us. Agents don't care about advertisements. Humans do. So if more of the utilitarian internet gets handed over to machines, where does human attention actually go β and which companies are positioned to monetize it? That leads us through Amazon vs. Walmart, Google vs. Meta, the economics of third-party sellers, YouTube advertising, and Dan making a sufficiently convincing case about Meta that Sean reluctantly changes his mind in real time. From there, we get increasingly agentic. Dan has been experimenting with TinCan, an open-source project that lets his various AI tools talk to one another, bringing us another step closer to the strange future where you don't manage individual AI applications so much as sit above an entire synthetic executive team and tell everyone what needs to happen. Then Sean shows up with a copy of Kurt Vonnegut's GalΓ‘pagos. Obviously. Sean has been spending an unreasonable amount of time turning his Tomes and Tunes record collection into a walkable 3D record store. There is no business model. There is no obvious problem being solved. It may, in fact, be completely pointless. Dan thinks asking why he's doing it is the wrong question. Just do things. That leads us to Vonnegut's blue-footed boobies, their bizarre courtship dance, and a surprisingly difficult question about AI and creativity: when does something stop being made by AI and start being made with AI? Maybe the distinction isn't quality. Maybe it's intent. An AI-generated bus advertisement doesn't need a soul. It has a job. But art is different. Art can be exploratory, purposeless, weird, and valuable precisely because somebody wanted to make the thing. We argue about AI watermarking, ballpoint pens that can suddenly write their own sentences, and whether today's obsession with separating "human" creation from "AI" creation can survive once the tools become ubiquitous. Or, as Dan puts it: eventually it becomes water. Finally, Sean unveils the first version of Sean's Manufacturing Index, an attempt to track whether the physical buildout happening across the American economy is beginning to show up in manufacturing labor markets. Manufacturing employment hasn't exploded, but manufacturing wage growth is currently running ahead of broader wage growth. Sean thinks labor churn, tariff-driven manufacturing-network changes, data centers, power infrastructure, switchgear, cooling systems, and other physical investment may be setting up an interesting labor-market inflection point. Which brings us, strangely enough, back to the original question. If AI makes certain kinds of digital labor abundant while the physical economy needs more electricians, technicians, builders, engineers, operators, and manufacturing workers... Where are the humans? We close by deciding that markets are memes all the way down, contemplating a Vibe Index, discovering an absurdly simple Mag 7 trading strategy, and wisely deciding to take the rest of that conversation off air. Probably for the best. β Sean & Dan Books Discussed GalΓ‘pagos by Kurt Vonnegut Unknown Market Wizards by Jack D. Schwager Shows/Films Discussed The Mighty Boosh Citizen Kane (and our pitch for a Christopher Nolan remake about Zuckerberg) Tools & Platforms Mentioned Muse (Meta's personal agent) TinCan / Agent Tincan (Matt Van Horn's open-source agent relay) Claude, ChatGPT / Codex, Gemini, Grok OpenClaw, Hermes (open personal-agent frameworks) Alexa (Amazon's shopping assistant) Sean's Manufacturing Index (our new SMI page) Sean's Tomes and Tunes (the 3D record store) Companies Discussed Meta, Google/Alphabet, YouTube, Amazon, AWS, Walmart, OpenAI, Anthropic, xAI, Oracle, Nike, Celsius, Coca-Cola, Budweiser, Miller Lite, Dove, Deckers, Skechers, Weber, June, Aalo Atomics, Jane Street, Colossus Links & References Sean's Tomes and Tunes: the 3D record store Sean's Manufacturing Index: the new SMI page GeekWire: Amazon blocks Meta's Muse AI assistant in a new standoff over agentic shopping Benzinga: Muse adds shopping with Walmart, Sephora and Best Buy (Meta Connect, Sep 23) Colossus: "Mark," Jeremy Stern's profile of Mark Zuckerberg Stratechery: Ben Thompson interviews Colossus EIC Jeremy Stern about the profile Agent Tincan: let your AI agents ask each other for help Sheel Mohnot on X: the "most hated" Mag 7 trading bot Marketplace Pulse: Amazon is regaining first-party unit share Dan's Substack, the Prometheus Dispatch Unqualified Fact-Check We said some things. Here's how we did. = Nailed it | = Close enough | = Whiffed it Meta released Muse Dan said Facebook just released Muse, its closed, just-works personal agent. It launched September 8 and hit #1 on Apple's US App Store within two weeks. Nailed it. Amazon blocked Muse this week, over ads Dan said Amazon blocked Muse from shopping on Amazon.com because agents don't care about ads. The block landed the weekend of September 20. Amazon's stated reasons were that the agent didn't identify itself and handled customer credentials. The subtext everyone reported was the $68 billion ad business, which depends on humans browsing sponsored listings. Right on both counts. Walmart has yet to do this Dan said Walmart hadn't blocked Muse. He undersold it. Walmart signed on as a Muse shopping partner at Meta Connect on September 23, alongside Sephora and Best Buy. Sean's "Walmart can be more AI-native than Amazon" was vindicated four days before we even said it. The eleven percent pop Dan wished he'd bought Meta right before its 11% jump. It happened on Monday, September 21: Wells Fargo raised its target from $640 to $796, and Muse was sitting at #1 in the App Store. Correct, and painful. From credit cards to wires Sean said some Amazon sellers now have to wire ad payments monthly instead of putting them on credit cards. Since April 15, 2026, sellers who still paid for ads by card have them deducted from their sales proceeds by default, with Pay by Invoice (net 30) as the alternative. Cards survive only as a backup. So the lost credit-card points are real, but the mechanism is deduction or invoice rather than a mandatory wire. Search is ninety percent, YouTube is ten Dan guessed search ads are about 90% of Google's revenue and YouTube ads about 10%, and flagged that he didn't know. YouTube is right at roughly 10% of Alphabet's $402.8 billion in 2025 revenue. Search and other is a little over half, not 90%, because Cloud, subscriptions and devices have grown into real businesses. Half credit for the hedge. Celsius is a pandemic-era brand Sean said Celsius was "pandemic era, I think." The company was founded in 2004 (Dan's guess that it had been around a while was closer). Its breakout was pandemic-era, though, capped by PepsiCo's $550 million stake in 2022. Right about the rise, wrong about the birth. Fun footnote: Celsius is one of the trades that made Chris Camillo's name. The composer who hated the phonograph Dan said one of America's more famous composers was loudly convinced the phonograph would ruin music. That's John Philip Sousa, whose 1906 essay "The Menace of Mechanical Music" warned that recordings would kill amateur music-making. Nailed it. BONUS: Sousa made plenty of records anyway. SMI at plus point six eight Sean said manufacturing wage growth is running about 0.68 percentage points ahead of other industries. August 2026 BLS data: manufacturing average hourly earnings +3.77% year over year against +3.09% for all private workers, a gap of 0.68. Exact, which is the benefit of reading off your own index. The 2009 spike Sean said the gap peaked around 2009β10 and went negative from 2010 to 2014, because manufacturing wages held steady while everything else fell. The timing checks out: the all-time high was April 2009 (+1.79 points), and it ran mostly negative from 2010 through 2013. The reason is mostly composition, not resilience. Layoffs take the lowest-paid workers first, which lifts the average. Sean's own SMI page says exactly that ("That is not a raise"), which makes this the rare fact-check where the host is corrected by himself. Twenty thousand jobs, five times the churn Sean said manufacturing added only about 20,000 jobs over the past year, with roughly five times that many moving between fields. Manufacturing employment went from 12.615 million to 12.638 million (+23,000). The five build-out industries on the SMI page gained 64,000 while the rest of manufacturing lost 41,000, which is 4.6 jobs moved for every net job added. One nuance: that measures jobs shifting between industries, not individual people changing jobs. Factory job openings up thirty percent Sean said factory job openings are up about 30% from a year ago. The JOLTS manufacturing job-openings rate rose from 3.3% to 4.4% (July 2025 to July 2026), up about a third. Nailed it. Four percent unemployment Dan asked whether unemployment was about 4% and admitted he hadn't looked. The August 2026 jobs report put it at 4.1%. Close enough for a guy who hadn't looked. Switchgear lead times Sean said switchgear lead times are running 18 to 24 months. Medium-voltage switchgear is quoting 52 to 80 weeks in 2026, and the 15β38kV gear that data centers need runs 80 to 104 weeks, about 18 to 24 months. That's roughly 2.5 times pre-pandemic lead times. Good, and more confident than the "I'm not enough of an electrical engineer" disclaimer suggested. One in four to one in six data center jobs stick Sean owned our Ep 54 miss and said one in four to one in six construction jobs stay on after a data center is built. The evidence is stingier than that. Meta's Hyperion campus peaks near 7,500 construction workers and settles around 1,000 permanent jobs (about 1 in 7.5). Stargate's figures run from 1 in 64 to about 1 in 6. Right direction, right humility, generous ratio. Solar vs. nuclear emissions Dan said solar produces more greenhouse gas than nuclear. On a lifecycle basis that's right: IPCC median estimates are about 41β48 gCO2e/kWh for solar PV against about 12 for nuclear. Both are a small fraction of coal or gas. Chris Camillo and the Deckers trade Dan said Chris Camillo, featured in Jack Schwager's Unknown Market Wizards , practiced "social arbitrage," reading trends off Twitter and buying Deckers when he saw Skechers taking off. The book, the chapter and the term are all right. Camillo's documented footwear trade was Crocs and UGG, and UGG is a Deckers brand. Right company, wrong sneaker. The worst-Mag-7 bot Dan said Sheel Mohnot built a bot that buys the worst-performing Mag 7 stock on a 30/60/90-day basis and heavily outperforms. It actually ranks on 3-, 6- and 12-month returns and rebalances monthly. Sheel reports +347% since May 2023, against +162% for the Mag 7 and +90% for the S&P 500. Wrong lookbacks, right result. At Sean's request we backtested it too (details in the episode analysis), and it holds up, with every hindsight-bias caveat you'd expect. BONUS IDs: "Sheel, starts with Monad" is Sheel Mohnot. And Sean's Aalo checks out: Aalo Atomics is an Austin microreactor company. And Matt Van Horn's "June stoves" was the June smart oven, acquired by Weber, which later shut down its servers. Dan was right about all of it. Final Score: 11 green, 7 yellow, 0 red A red-free week, and the one host who corrected himself did it on his own website. We'll take it. Chapters 00:00 - Cold Open: Where Are the Humans? 01:34 - Meta Launches Muse 04:55 - Amazon Blocks Muse: Agents Don't Care About Ads 05:37 - Walmart Can Be More AI-Native Than Amazon 11:50 - Stock Face-Off: Meta or Google for Ten Years? 16:33 - YouTube, Brand Spend and Celsius 19:09 - TinCan: Getting All Your Agents Talking 23:02 - Tomes and Tunes Becomes a 3D Record Store 25:24 - GalΓ‘pagos and the Blue-Footed Booby 27:08 - Made With AI vs. Made By AI 33:18 - Credit the Writer, Not the Ballpoint Pen 35:09 - Back to Ep 54: Do Data Center Jobs Stick? 37:08 - Introducing Sean's Manufacturing Index 46:38 - The Upper Middle Class Gets Walloped 49:45 - Vibe Investors and Social Arbitrage 52:17 - The Worst-Mag-7 Trading Bot






