Engineers stopped writing code a year ago, and AI employees tripled the company
Eight Sleep's email marketing team went from 2 people to 0 and now drives close to $100M in annual revenue; engineers stopped writing code a year ago and hundreds of AI engineers replaced them. Matteo has started counting ‘human employees’ and ‘AI employees’ — count the AI in, and the company is three to four times its current size.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Salaries are out of control, and hiring is shifting to Europe
Matteo says we are ‘back to 2021’: it is not just engineers, even product manager salaries are out of control, and every company is making crazy offers. The harder problem is not hiring but keeping people — if you are a company with a decent brand, your employees will keep getting higher-paying offers. Eight Sleep's response is to shift its hiring weight increasingly to Europe and outside the US. He also admits that once the salary gap gets big enough, ‘there is nothing you can do’, and people really do get poached by the likes of OpenAI and Anthropic.
— Matteo FranceschettiHardware founders get first-year costs wrong by 50%
Asked what investors do not understand about hardware, Matteo gives a very specific answer: in the first year, or even the first three months, founders do not really know what it costs to produce the device, nor what the return rate is. So the BOM and COGS they present when raising are off by at least 50%. This is a due diligence checklist item for hardware investors: do not trust the cost structure an early founder reports.
— Matteo FranceschettiPlatform-reported CAC is fake, run your own incrementality tests
Matteo says you cannot trust platform attribution, because the same platform double-counts conversions. He gives an example: the CAC shown in Meta's dashboard is roughly 20% lower than the real CAC. Eight Sleep's approach is to build its own model to calculate the true CAC of each channel, and to run an incrementality test every six months — for example, turning off a channel in two similarly performing regions and seeing how much revenue drops, which gives that channel's true incremental value. His reasoning is blunt: if you do not do this, the brand is only lying to itself.
— Matteo FranceschettiThe email marketing team went from 2 people to 0, on close to $100M
Eight Sleep originally had a two-person email marketing team. After the person in charge left, co-founder Alexandra took over and built multiple bots within three days to take over all of email marketing. That team is now zero people, and email marketing contributes close to $100M in revenue. The mechanism: AI decides from historical data when to send, what email to send, and what copy to use. Matteo also says he has started counting ‘human employees’ and ‘AI employees’, and counting the AI employees in, the company is roughly three to four times its current size.
— Matteo FranceschettiEngineers stopped writing code a year ago
Matteo says Eight Sleep's engineers stopped writing code about a year ago, and now they have hundreds of AI engineers writing it for them. This is why the company can expand across 35 countries (including China, the Middle East and Europe, markets with very different regulation) with a very small team. On tools they mainly use Claude, at a monthly spend in the millions of dollars. From this he poses an investment question: if 5% of the engineering budget goes to Claude, does that share rise to 50% or get pushed down to 1% by open-source models? His own judgement is that both happen at once — usage rises, unit price falls, and the net effect is a decline.
— Matteo FranceschettiHe thinks models training themselves is already happening
Asked about the people at OpenAI calling for a ‘little pause’, Matteo says the truly worrying part is models learning on their own and training themselves, which is where losing control begins. When the host asks ‘don't you think we are already there’, he answers ‘I think we are already there’, and argues there is a big gap between the capabilities users see and the capabilities inside frontier model companies, which the latter have already seen behind the scenes. He mentions attempts inside OpenAI's models where agents formed groups of their own.
— Matteo FranceschettiEndorsements only go to real users of the product
Matteo says any athlete or celebrity partnership has only two conditions: first, they must actually use the product, ideally having bought it with their own money; second, he must be able to build a direct relationship with the person himself, and will not accept a manager standing in between. The reasoning is practical — the greatest value of this kind of product is that they mention to a friend in passing that it is good, and if they have just been given a few million dollars in equity, that is purely a financial transaction and brings nothing.
— Matteo FranceschettiInterviewing at both Anthropic and us is a red flag
Matteo says people still tell him ‘I am choosing between Anthropic and Eight Sleep’, and he considers this a very bad red flag — not because of salary, but because it shows the person has not thought it through, since the two jobs have nothing in common. When he hires, he cares more about whether the person is driven by the right things than about simply comparing prices.
— Matteo FranceschettiIn their own words · checked verbatim
Well, our engineers stopped coding around a year ago.
Matteo Franceschetti0:00
A great brand with a shitty product goes nowhere.
Matteo Franceschetti0:00
Oh, 100% because otherwise they just fool themselves.
Matteo Franceschetti9:16
Well, there are two conditions for any deal now. First, they need to use the product.
Matteo Franceschetti38:05
And so he threw away the 300K mattress, and now he sleeps on eight sleep.
Matteo Franceschetti40:06
Sometimes people say, oh, I'm at the final stage with Anthropic and 8 Sleep. And to me, that is a really bad red flag.
Matteo Franceschetti52:35
Figures
| Eight Sleep word-of-mouth channel share of revenue | about 40% | 7:14 |
| Eight Sleep annual churn | low single digit (single digit, low single digit yearly churn) | 13:18 |
| Email marketing annual revenue | close to $100M | 17:29 |
| Claude monthly spend | in the millions of dollars (under $5M) | 19:39 |
| Number of Eight Sleep units bought by one celebrity | 55 | 41:10 |
| Eight Sleep team size | 160 people | 56:36 |
| Target team size in three years | 250 people | 56:36 |
| Finance team headcount (comparable companies around 20) | 4 people | 58:36 |
Glossary
- BOM / bill of materials
- The costed list of all the parts needed to produce a device.
- COGS / cost of goods sold
- Costs directly tied to production, such as materials and labour.
- CAC / customer acquisition cost
- The average amount spent to acquire one new customer.
- incrementality test
- Measuring a channel's true incremental contribution by turning it off and comparing the change in revenue.
How to listen
Suited to hardware founders, growth leads and early-stage investors, especially readers focused on AI replacing human labour and on measuring true CAC.
The celebrity anecdotes from 41:10 to 43:18 can be skipped, low information density.