Matteo Franceschetti on AI Replacing Teams at Eight Sleep

A year ago, the engineers at Eight Sleep stopped writing code. Not because the company stopped building things, but because it built something else first: an internal fleet of AI agents that now write the code while people supervise. Matteo Franceschetti, co-founder of the sleep tech company Eight Sleep, mentions this almost as an aside on The Twenty Minute VC with Harry Stebbings. It turns out to be the clearest thread through an unusually candid conversation about what happens to a company's org chart once intelligence gets cheap.

Eight Sleep sells a sleep-tracking cover that fits over a mattress and connects to a subscription service on top of the hardware. The company now sells in 35 countries, including China, and Franceschetti has done this show before, which is partly why the second conversation runs looser and more specific than the first.

The clearest evidence for the coding claim is not the claim itself. It is what happened to the email marketing team. The employee running it quit (17:10). Rather than hire a replacement, Franceschetti's co-founder and wife, Alexandra Zatarain, spent three days building bots, using the AI model Claude, that now handle the entire email calendar: when to send an email, to whom, and with what copy, based on historical performance data (17:21). The channel now runs with a human headcount of zero and still produces close to $100 million in revenue (17:41).

"Our engineers stopped coding around a year ago." — Matteo Franceschetti [19:42]

The same logic runs through the rest of the company. Eight Sleep has 160 employees total (56:47). Its paid media team, which manages hundreds of millions of dollars in ad spend, is two people (57:10). Its finance team is four people, where Franceschetti estimates a comparable company at Eight Sleep's revenue would run a team of twenty (58:45). He describes his organizing principle as building "teams of two": two people, ideally excellent, given responsibility for an area that would normally require a department, with AI agents doing the layer of work in between. The result, he says, is revenue per employee higher than Apple's (56:47).

Behind those small teams sits a group that did not exist six months before this conversation: an internal AI tools team, built to construct and maintain the thousands of agents used across the company. Franceschetti keeps it deliberately close to the data engineering team, since every marketing or finance agent needs reliable access to the same backend numbers, whether that is revenue by country or cost of acquiring a customer by channel (58:01).

The Discipline Behind the Story

It would be easy to read all this as an AI hype story, but the more interesting part of the conversation is how unglamorous the underlying discipline is. Franceschetti says hardware founders reliably underestimate their bill of materials, the itemized cost of every component in a product, and their total cost of goods sold by at least 50% in the first year (07:25). He also treats word of mouth as a metric to track rather than a vague compliment: organic word of mouth drives about 40% of Eight Sleep's revenue even at its current scale (08:06), and he says this is one of the first things he now checks as an angel investor sizing up other consumer companies.

On paid marketing, Franceschetti's rule is to scale one channel at a time. Start with the largest platform, usually Meta, set a hard cap on cost-per-acquisition, and let the team spend as much as it wants as long as it stays under that cap (10:50). Every six months or so, the company runs an incrementality test: it turns a channel off in one region, such as one state, while leaving it on in a comparable region, and compares the growth rates to isolate what that channel is actually worth (09:10). The reason is that platform-reported cost of acquisition cannot be trusted on its own; Franceschetti says Meta's own attribution understates the true cost by about 20% (09:30).

Brand building follows a similar rule: prove the relationship before signing a check. Eight Sleep's partnership with the Formula One driver Charles Leclerc began years before it was a formal deal, when Franceschetti, a lifelong motorsport fan, tried to sponsor Leclerc while he was still racing in Formula Two, long before Eight Sleep had any real money to offer (35:32). By the time the two actually worked together, Leclerc had already been sleeping on the product for two years without either side knowing it. Franceschetti says roughly 80% of Formula One drivers now sleep on an Eight Sleep product, though most of those relationships stay private (39:18). His rule for any athlete deal is the same: the person has to already use and like the product, and he insists on a direct relationship rather than one filtered through a manager, since otherwise, in his telling, it just becomes a financial transaction that never turns into real word of mouth (38:04).

That same principle, proof before commitment, carries into how Eight Sleep entered a market with almost no overlap with how it sells everywhere else.

"Chinese don't buy on websites. There is no e-commerce as we know it." — Matteo Franceschetti [28:42]

Chinese consumers buy almost everything through super-apps such as WeChat and RedNote rather than through a company's own site, which meant Eight Sleep had to rebuild its entire go-to-market approach from nothing (28:20). Franceschetti also argues the country's manufacturers, having already out-designed European automakers on electric vehicles, are positioned to do the same with humanoid robots, a market he compares to what cars represented in the 1960s (25:36).

What Happens When Intelligence Is Free

The boldest part of the conversation is not about Eight Sleep at all. Franceschetti predicts that children born today will never need to work for money, because by the time they are adults, AI will have automated most of the jobs that currently support households (60:12). He does not treat this as pure upside. He expects taxation to shift toward the companies whose margins improve most from AI, to fund something like a universal basic income (61:50), and he calls university "100% useless" for anyone planning a career as a Silicon Valley entrepreneur (61:50), a claim that says more about the founder culture he operates in than about universities generally.

He is more measured about nearer-term shifts: a personal AI agent one day buying and operating a self-driving robotaxi on its owner's behalf, the way his own AI assistant already books errands for him (30:22); the slow normalization of letting AI systems hold financial information, which he compares to the discomfort people felt entering a credit card number online two decades ago (29:36). None of it reads as certainty. It reads as a founder thinking out loud about the edges of a system he is already living inside, where the same tools that eliminated his email marketing team are quietly rewriting what a company, and eventually a career, is for.

Near the end, Franceschetti offers a plainer frame for all of it: a pyramid with health at the base, relationships in the middle, and purpose at the top, in that order, because none of the higher layers matter if the one below it is missing (60:53). It is a modest thing to say after forty minutes of predictions about robots and automated labor, but it is also the quiet argument underneath the whole interview. The technology can take over the coding, the emails, and eventually the driving. What is left, in his telling, is the part of a job that was never really about output at all.