Summary

Matteo Franceschetti says the talent market is “back to 2021,” with salaries “out of control” even for product managers — so Eight Sleep is shifting hiring to Europe and outside the US. The real problem isn’t attracting people “on T zero” but retention against relentless re-offers; he’s lost “a couple” of people to hot AI companies where “the delta in salary and in comp is so big that at that point there is nothing you can do.”
Eight Sleep’s growth machine runs on ruthless channel discipline: one channel at a time, hard CAC caps, and six-monthly incrementality tests, because platform attribution overstates performance. Meta’s reported CAC runs ~20% below true CAC, so Eight Sleep built its own models; word of mouth compounds to ~40% of revenue, churn is “low single digit” yearly — better than wearables going public — and “we could be growing 50% more if we wanted… but then next year you’re in trouble because your CAC is upside down.”
“Our engineers stopped coding around a year ago” — they direct hundreds of AI engineers mainly on Claude, spend “in the millions,” and email marketing now makes close to $100M with a team of zero. Counting “AI employees,” the 160-person company is “probably 3, 4X bigger”; revenue per employee is “way higher than Apple,” paid media makes $200–300M with two people, and the three-year target is “250 people making a billion.”
On Harry’s Anthropic IPO question — Harry cites Marc Benioff saying “they” spend $300M/year on Claude against a $6B engineering budget, so does that 5% go to 25% or 1%? Matteo sees “two vectors in the opposite direction.” Usage takes 5% to 50% while cost collapses, “so net-net I think it will go down,” echoing Sam Altman’s electricity analogy; he’d still “probably buy” the IPO and is personally most bullish on SpaceX.
His industrial thesis: AI turns the right companies into Xiaomi-style holdings owning “3, 4, 5, 10 different businesses,” and automakers must pivot to humanoids “because the humanoid will be what cars were in the ’60s.” Chinese EVs already have “Ferrari aesthetics” and Apple-store retail; at a robot competition Harry called the Humanoid Games, a robot reportedly ran the 100 meters faster than Usain Bolt, and on model self-learning — the point “when we start losing control” — asked if we’re already there: “I think we are.”
China commerce and agentic buying are converging: “Chinese don’t buy on websites” — purchases happen through super apps like WeChat and RedNote — while his Grok bot already shops via Amazon on the browser, which holds the card. His crazy-today-normal-in-five-years call: “your bot buying a cyber cab that will start making money for you… the bot runs the cyber cab” — and the Eight Sleep bed becoming “the most advanced medical platform you ever had,” with the company starting to see diabetes and hypertension likelihood in its data.
The labor predictions are categorical: “kids that are being born now will never work, at least for money”; UBI “will have to happen” via taxing more-efficient companies’ improving bottom lines; university is “100% no.” It’s useless apart from the social part. The matching talent red flag: candidates at “the final stage with Anthropic and Eight Sleep” signal confusion — “there is no way that these two jobs are in any way similar.”

Deep dive

1. Hiring is “back to 2021” — and the real problem is retention

Matteo’s read on the talent market: “We are back to 2021,” with salaries “out of control” — not just engineers but “even for product managers” — as substantial amounts of money circulate through offers. Eight Sleep’s structural response is to hire more and more in Europe and outside the US.
The sharper point: the bottleneck isn’t attracting people “on T zero,” it’s retention — “if you are a cool company with a cool brand, then they keep receiving offers over and over at a higher salary.” He’s lost “a couple” to OpenAI, Anthropic, or other AI companies doing extremely well or that have just raised money; sometimes “the delta in salary and in comp is so big that at that point there is nothing you can do.”

2. Angel investing builds founder intelligence; hardware founders’ costs are off by 50%

Why he angels — a lesson he says isn’t shared enough: “you build your own intelligence.” Investor updates from non-competing companies reveal valuations, multiples, hiring profiles, and which channels and markets might be working.
His warning to Harry-as-hardware-investor: in the first months, founders “don’t fully understand what the cost of producing that device will be… everything they tell you about BOM and COGS will probably be off by at least 50%,” and they can’t yet see return rates.
The growth variable that matters is word of mouth — in Eight Sleep’s case it keeps compounding and drives ~40% of revenue “even at our size.” Paid media (Meta, Google, TikTok) comes next.

3. One channel at a time, CAC caps, and never trust the platform’s attribution

The playbook he gave a mutual friend asking whether to put $100K into every channel: “No, no, no. You go one by one.” Start with Meta, set a weekly budget and a hard CAC cap; teams can deploy as much as they can within that cap — “$100 million if they are within the cap of the CAC” — but only while inside it.
Every six months, run incrementality tests: turn a channel off in one geo (his Texas/California example) and measure the revenue delta versus the comparable control. Meta’s reported CAC is “probably 20% lower than what the true CAC is,” which is why Eight Sleep built its own models — every brand needs them, “100%… otherwise they just fool themselves.”
The discipline is about next year, not this one: growth people always want to spend more, and “we could be growing 50% more year over year this year just by spending more. But then next year you’re in trouble because your CAC is upside down.” Harry’s gloss: good luck fundraising going from 400% growth to 70%.

4. Payback on day one, churn that’s “world-class,” and a new report line called AI SEO

Keith Rabois “was really pushy, and he was correct from day one” on immediate payback: healthy contribution profit on day zero, with the subscription compounding on top. Compressing it would allow 50% more growth, but the business “wouldn’t be public-ready or as healthy as it is today.”
Subscription churn is “low single digit yearly” — “the lowest of any company you can think of,” better than wearable companies going public (“Who could that be? Yeah.”). His twice-weekly report now includes AI SEO — ranking inside AI searches. He couldn’t give a percentage for AI-search traffic; when Harry pressed him between 1% and 20%, he said it was “probably somewhere in the middle.” Their assumption is that Google is behaving differently because of AI searches.
Channel experiments as told: out-of-home in the Middle East was his biggest mistake — too early, since awareness must precede it. Illustratively, he said that if 100 people see an ad, perhaps 5% might know the brand. TikTok finally worked six months ago via influencer UGC volume; and a bookable truck with a bedroom built into the back now roams Silicon Valley, letting people try the product and buy it.

5. Teams of zero and teams of two: 160 people out-producing thousand-person peers

The forced experiment: when the email lead quit, co-founder Alexandra “within three days was able to build multiple bots that now run all our email marketing.” Team of zero; email makes “close to 100 million,” with AI using historical data to suggest when to send which email and with what copy.
He now counts “human employees” and “AI employees” — hundreds if not thousands of internal agents — “we are probably 3, 4X bigger if you start including all our AI employees.” Paid media agents deliver morning change recommendations humans approve or reject; that two-person team makes $200–300M, finance runs on four people instead of a team that might otherwise be around 20, and a six-month-old AI internal-tools team wires agents into the data backend and pipelines so they can be kept accurate.
The org math: 160 people with revenue per employee “way higher than Apple’s” — comparable companies carry ~1,000 heads at his revenue stage, and “our output is 5X the average in Silicon Valley.” Three years out: “250 people making a billion,” organized as teams of two world-class people with redundancy over huge areas. Nothing disappears; people spread horizontally.
On infinite AI content drowning discovery, his answer is agent-side optimization: “the personal agents will look for the best product, and so if you are the best product you will still win” — one strategy for showing up where humans go, and a different one so agents find and recommend you.

6. “Our engineers stopped coding around a year ago” — and the Anthropic spend math

Engineers now direct “hundreds of AI engineers that code for them,” mainly on Claude — spend is “in the millions,” growing so fast he doesn’t track it — which is how a fairly tiny team operates at scale in 35 countries including China, the Middle East, and Europe, all with substantial regulation.
Harry’s investor puzzle cites Marc Benioff’s numbers: $300M/year on Claude against a $6B engineering budget = 5%. Does 5% go to 25% — “load up on Anthropic’s IPO” — or does open-model commoditization take it to 1%? Matteo: “two vectors in the opposite direction. The usage will increase, and so the 5% will become 50%, but the cost will go down. And so net-net I think it will go down” — citing Sam Altman’s electricity analogy: early on, paying for one hour of electricity at night required five hours of work; now nobody knows the price.
He’d still “probably buy” Anthropic at IPO — “there is so much upside in all these companies” — but he’s personally most bullish on SpaceX.

7. AI turns companies into holdings, and automakers must become humanoid companies

The holdings theory: with AI, “the right companies will transform themselves into holdings” — Eight Sleep Inc. eventually owning “3, 4, 5, 10 different businesses” as internal tools reveal adjacent opportunities. Chinese companies prove the model to him: Xiaomi “does 15 different things” including cars; BYD develops components beyond the car and automotive space. “The concept of focus is not, oh, you keep doing the same thing forever.”
On Chinese autos: around 10 years ago they chose not to fight European manufacturers on standard engines and doubled down on electric; now Xiaomi stores look like Apple stores with cars — “Ferrari aesthetics,” rotating seats, everything voice-controlled — while other manufacturers “don’t focus enough on the user experience.”
His bigger worry than Europe’s car industry: “all these automakers should transform themselves and start building humanoids, because the humanoid will be what cars were in the ’60s” — miss it and they lose both cars and a trillions-of-dollars humanoid market. The robot competition Harry called the Humanoid Games looks clunky, but Matteo says a robot ran the 100 meters faster than Usain Bolt, and “in one year from now, we will be blown away” by AI’s speed.
On a paper, essay, or post from someone at OpenAI that Sam Altman promoted: the frightening part is self-learning — models “train themselves, and that is when we start losing control.” Harry asked whether we’re already there; Matteo answered, “I think we are,” with a “big delta” between what users see and what frontier labs “have already seen behind the scenes.”

8. China has no e-commerce websites — and your bot will buy a Cybercab

Selling in China meant relearning everything: “Chinese don’t buy on websites. There is no e-commerce as we know it.” Purchases happen inside super apps — WeChat, RedNote and equivalents — where their credit cards are used; websites are mainly for scrolling through pictures and reading about products. It’s also why China is his hardest of 35 countries.
On agent payment trust: “my Grok bot is buying things for me already, but it’s using Amazon on the browser, and so Amazon has my credit card.” We’re where internet credit cards were 20 years ago — Italians assumed many sites were a scam — and within a year or two, he expects people to start trusting bots from frontier models such as X, OpenAI, and Claude.
His crazy-today prediction: “your bot buying a cyber cab that will start making money for you… the bot runs the cyber cab.” On health, he hopes cancer is solved within 10 years; Eight Sleep is starting to see in its data the likelihood of developing diabetes and hypertension and says it is becoming a sleep-apnea medical device — “your bed will become the most advanced medical platform you ever had.”

9. Timing beats binary, and brand beats features

The CEO lesson: channels and markets aren’t binary works/doesn’t — “there is a time for things.” Retail becomes a meaningful Eight Sleep channel “in the next months and years”; earlier would have been too early.
The China-entry logic, learned from Chinese founders who launch abroad first: “the Chinese market loves to buy products that are already well-recognized in the Western world.” Entering five years ago, “we would have been destroyed by a knockoff that is 500 bucks cheaper”; now Eight Sleep sells in 35 countries, sponsors Charles Leclerc and works with Ferrari through him, is used by many athletes, and has been discussed by Elon Musk and Mark Zuckerberg. Matteo says that level of trust lets the brand command its price.
On Hugo Barra’s price-king-or-feature-king rule, he agrees and disagrees: the winning axis isn’t features but product plus brand. His specimen is the iPod — Samsung had 1,000 songs in a clunky device too, but Jobs arrived “sometimes one, two, three years later” and “was nailing the branding, the packaging, and the pitch.” Silicon Valley’s blind spot: “we understand really well building the best product, but we underestimate the importance of brand. Then a great brand with a shitty product goes nowhere.”

10. The athlete playbook: they must already use it, and no managers in the middle

The Leclerc story: Charles had slept on Eight Sleep for two years before they met — and Matteo still had the emails from trying to sponsor him in F2, before F1, when Eight Sleep was too tiny to close a deal. At a dinner he thinks was on the Wednesday of the Miami GP, the deal was done in days — and Leclerc invested.
Two non-negotiables for any deal: they must genuinely use the product (“otherwise it’s just a pure business transaction and I don’t like it”) and Matteo must have a direct relationship. If managers stay in the middle and he lacks that relationship, he won’t do the deal, because “the biggest value they provide is simply talking to their friends about how good our product is.”
The undisclosables: ~80% of F1 drivers sleep on it, top tennis players ping him at Grand Slams; a famous footballer threw away a custom $300K mattress to move Eight Sleep into his bedroom; one famous person owns 55 units for his houses, embassies have bought in the hundreds — and “these people don’t ask for discount.” On premium positioning: “imagine if we were Tesla, we are the Model S” — other models are coming to “democratize sleep.”

11. Slack for work, WhatsApp for marriage — and how to keep people from the labs

Running the company with his wife Alexandra: two channels, never mixed — screaming on Slack about a 9 PM review, pizza cravings on WhatsApp. Her framing that fixed the dynamic: “If I were an executive who wasn’t sitting next to you, would you call them for this thing at 10:15 PM?… Treat me how you would treat our VP of Operations who is in New York.” Dalian and Keith, Matteo thinks, didn’t realize they were married during their first year of board meetings.
The retention war story: his hardware lead called at 2 AM on a Friday during a Maldives weekend to say he was leaving for a company offering liquid RSUs and probably roughly 3X the pay. Matteo spent the weekend trying to save him; five minutes before the employee was due to tell his team he was leaving, he asked Eight Sleep for an offer, and stayed. Matteo attributes the decision to the employee’s love of the company’s culture and work, alongside the improved offer.
The red flag: “Sometimes people say, ‘Oh, I’m at the final stage with Anthropic and Eight Sleep.’ And to me, that is a really bad red flag… it means you’re confused. There is no way that these two jobs are in any way similar.” A newer Valley pattern in his data: laid-off candidates accept offers, then switch companies two weeks later, apparently to restart income while keeping other conversations open.
Before promoting his internal VP of Engineering candidate — who proved himself over three months while an external search also ran — Matteo refused a same-day answer: “Go and talk to your wife… both of you need to commit, because it’s not just you. It’s your family.”

12. “Kids born now will never work” — and the drive he can’t explain

The labor-market predictions, stated flat: “kids that are being born now will never work, at least for money” — by the time they are 20, he expects AI to have reached a point where humans do not work, though they may pursue activities and passions. UBI “will have to happen,” funded by taxing companies whose bottom lines become more efficient through AI. University for an aspiring Silicon Valley entrepreneur: “100% no. It’s useless” — apart from the social part.
What replaces work is his pyramid of life: health at the base (“if you have cancer… you can’t be happy”), then relationships, then purpose — “you need to work toward something… that is the part that we need to pay attention to.”
The origin as told: Ferrara, 150,000 people, walls from the 1600s; no English until 24; averaging 23/30 in law school until a magazine associated with Il Sole 24 Ore set two conditions for international law firms — English and magna cum laude — and “in my typical fashion, I say, okay, I’ll get it done”: near-perfect scores, his dad’s loan to study English in San Diego, Freshfields, Allen & Overy. His father died while he was still a Milan lawyer — “I’m the same person but in two different worlds” — and part of his drive is “probably” something to meet that bar. He’s now interviewing the parents of successful athletes and other high achievers as due diligence on “how I can be a dad one day.”
The honest weakness he and Alexandra call “the scene”: no vacation since Eight Sleep started, no honeymoon, bored anywhere after two nights. Who he admires: a man dining alone in Milan, perfectly dressed, no phone, “just thinking and enjoying his meal in this, apparently to me, peace. That is something I struggle with.”

AI Kills Everybody or Doomer Psyop? OpenAI’s Math Breakthrough, Nike’s $200B Collapse