Elad Gil: Silicon Valley’s Most Dangerous Startup Advice

Elad Gil takes apart Silicon Valley's standard startup advice

Silicon Valley's advice about how to start a company is inherited rather than tested. Gil argues the cofounder rule came from Y Combinator's equal splits, that winning rather than perks makes a culture, and that this moment is unusual because technology and buying behaviour are shifting together, so customers will try things they normally refuse. That is why buying a business and expanding its margins with AI now counts as a path to a company.

Takeaways

• The cofounder rule is mostly wrong. Gil says the rule that you always need a cofounder is mostly wrong. Michael Dell and Larry Ellison had none, Steve Jobs was the dominant founder of two, and Y Combinator normalised the equal split.
• Customers now say yes to new tools. Customers are far more willing to try new things because every CEO now has an AI edict, and building is dramatically faster. Gil says markets usually open on a technology shift or a change in buying behaviour, and both are happening now.
• AI roll-ups need buying and operations. One path at the pre-idea stage is buying existing businesses and expanding their margins with AI. That takes the ability to buy and the ability to operationally change what you buy, with AI applied on top.
• His model oligopoly call was half right. Gil predicted an oligopoly of foundation labs aligned to hyperscalers, and got that right. He missed Meta and the Chinese government funding open source. Stickiness now comes from the harness as much as the model, since few users switched when a new OpenAI model shipped.
• Durability comes from owning the data. Applications last when they are the system of record, holding the core data about a thing and its attributes, as Workday does for a company's people. A dozen integrated products cross-sold to one customer works too, as Rippling and HubSpot show.
• Winning, not perks, makes a culture. Gil says winning is the biggest determinant of culture, not perks, and companies that work tend to work early. The failure modes are sticking too long with something that isn't working and taking advice from people who have never had anything work.
• Raise size should match what you spend on. How much to raise depends on the goal and on the split between GPUs and headcount. Gil sees companies putting 80 percent toward GPU. Agarwal argues a $100m seed only works if everything goes right for 12 to 18 months.

What Gil expects next

• Crypto gets worse before it gets better. Gil stays a long-term bull but puts Bitcoin at $35,000 to $40,000, and calls Stripe the hidden crypto company.
• Many AI drug discovery companies end up as ordinary drug developers. A drug costs about $1.5bn and takes fifteen years.