Barrels, Undiscovered Talent and Ugly Babies: Keith Rabois on Building Companies in the AI Era

Rabois starts from a lesson Vinod Khosla gave him at Square: the team you build is the company you build. Markets, customers and technology distract founders from what really matters. He traces PayPal's lasting influence to Peter Thiel and Max Levchin assembling an unusual density of talent, largely through their networks.

He admits he was mediocre at hiring early on, roughly 50/50, and got feedback from David Sacks that he needed to show leadership leverage, with each hire returning more than one. He responded by recruiting people already inside PayPal whose talent was underused, and was promoted quickly. The lesson he took was that he could assess people he knew well but not strangers in a 20 to 45 minute interview.

Rabois says a founder who can ruthlessly and accurately assess talent can go far with no other abilities. He compares hiring to a muscle and notes tactics can move you within the middle of the bell curve but cannot teach you to find the top basis points. One teachable tactic is exhaustive referencing: he says DoorDash's Tony does 20 references on every senior hire, and David Sze's rule was not to stop until you hit a negative reference.

Framing matters. He says many investors passed on Fair because they asked whether its founder Max Rhodes was a good employee at Square rather than whether he could be a world-class entrepreneur. For senior candidates, Rabois asks what they would have done differently as CEO of their company, then probes why they could not persuade the actual CEO. He also recommends asking whether you would make the same decision 30 days after a hire, which he says predicts outcomes about as well as waiting a year or two.

Rabois describes a recurring conversation with CEOs who raised money, hired many people and then found that output per unit of time had not improved. His explanation is that the number of people who can independently drive an initiative from inception to success is very limited. He calls them barrels, and hiring more people without more barrels just stacks people behind the same initiatives and raises the collaboration and coordination tax.

He says PayPal had about 254 people in Mountain View when it was acquired and, by his strict count, 12 to 17 barrels. He recalls Jack Altman saying Lotus had two. The ratio of barrels to ammunition dictates how many important initiatives can run in parallel, and he stresses that ammunition is still needed, in amounts that depend on the project.

A barrel, he says, is someone to whom you can point at a hill and say get us over it, and they will motivate people, gather resources and measure what's needed. You can fire and forget, and if it will not work they come back early with the diagnosis and what they have tried. Rabois says the word agency is accurate but easily goes in one ear and out the other.

His example is the smoothie test. He wanted cold, healthy smoothies delivered to Square's engineers at 9 p.m., and the office team failed repeatedly. On his second day, an intern named Taylor Francis said he would solve it, and the smoothies arrived cold and on time. Rabois says he realized he had found a barrel and later gave the intern almost everything to do.

On attracting top people, Rabois says mission and vision are still indispensable, but a more nuanced approach is to show a candidate that their specific skill overlaps with the company's biggest blocker. If marketing is the constraint, a world-class marketer is betting on themselves. He says this argument, that he was one of only two or three people in the world who understood financial services and was still entrepreneurial, is how Square got him to leave Google after two weeks.

His bigger theme is to build on undiscovered talent, something he says Peter Thiel taught him in his first week at PayPal. Startups have maybe a tenth of the compensation budget of the incumbents they compete with, like a salary cap, so competing for universally wanted people is a losing game. He points listeners to his hiring talk at Ramp.

To spot undiscovered talent, he asks why large recruiting machines would misprocess a person: what would Meta, Google or Coinbase miss about them, and why? Sometimes the answer is a lack of information. That is why he says this approach skews younger, as young candidates have few data points, like a thin credit file, so a homogeneous evaluation process cannot score them. Lenny notes the skill matches how investors pick startups.

Colleagues describe Rabois as a bar raiser who pushes harder as the numbers improve. He relays Mike Moritz's answer that the common denominator of the best CEOs is the relentless application of force. Success breeds complacency, he argues, and most businesses are not network-effect businesses that can run on their own, so offsetting complacency is the CEO's single role.

He adapts his feedback to context. When a company is struggling, he is usually non-critical, more of a coach and supporter, because the founder already knows. When it is thriving, he is critical, isolating future problems while everyone is happy and borderline complacent. He likens it to a sports coach who polishes details while winning and sells the future while losing.

He likens a good year to winning a Super Bowl: the record resets to zero the next season. In venture, he says, he is only as good as his last investment and has to find another undiscovered founder every day. Lenny notes that at Airbnb, breaks sometimes lowered morale, and Rabois agrees that talented people are like superb athletes who dislike it when others coast.

Asked for career advice amid AI anxiety, Rabois says AI will radically reorient many careers, including his, and the way to thrive is intellectual curiosity. He says that in some of the best organizations he sees, the number one consumer of tokens is the CMO. Executives there build analytics and even campaigns directly rather than relying on layers of deputies.

Rabois cites a podcast with Peter Fenton that convinced him that the conventional PM makes little sense, since PMs take customer inputs and build a sequential year-long roadmap. Model capabilities change so quickly that something impossible in November may be easy by March. The winning startup, he says, notices what is newly possible this week and ships new value next week.

He and Lenny agree that as building gets easier, the scarce skill is deciding what to build and aligning people around it. In his view whether you were a PM, designer or engineer, the skill now resembles being a CEO. He cites Alfred Lin's chef analogy: a famous chef is mostly defining the value proposition, brand and pricing, not cooking each dish.

Rabois says many great engineers have always had commercial instincts, naming Max Levchin and Jeremy Stoppelman. He expects AI to put a premium on engineers with business sense, since they will not need large teams. He describes a Ramp engineering director who manages about 20 people yet ships as much code as he did as an individual contributor, using AI like a second team.

Rabois says design and code are merging, and it is unclear whether code becomes design or design translates into code. He gives Shopify as a concrete example: for over two years, he says, PMs there have not been allowed to present PowerPoint or Keynote decks. Every product presentation must be a working demo. Lenny adds that open design roles have plateaued over three years.

Even so, Rabois believes the alpha is in design, as in marketing, because it is about storytelling rather than tools or channels. The person who can frame a concept in the snappiest, most compelling way cuts through the clutter, which he says is worth more than all the tools. Lenny suggests AI is eating the middle of software development, leaving what to build at one end and distribution at the other.

Asked whether AI content will surpass human content, he calls it inevitable, but expects two poles: a curated, premium-provenance human tier, like a Warhol, and an algorithmic ranking of whatever is best. He attributes AI writing's weakness partly to token rationing by model providers, saying short outputs are better than long ones, so he favors many short examples and editing.

Drawing on his earlier career as a litigator, he says the magic of a brief was its first paragraph, and he might spend a week or more thinking about the first three sentences before writing 30 pages in two days. He also says legal training is largely at odds with entrepreneurship: law school rewards issue spotting, while founders must solve problems. Billing by the hour also took him a year and a half to stop tracking his time.

Rabois says he hates talking to customers and refuses to let colleagues do it for consumer, SMB or micro-merchant products. Customers answer subconscious decisions consciously, so even sincere answers mislead. His example is that someone who bought a Porsche will give you every reason except the real one. In enterprise, where a decision maker makes mostly utilitarian choices, such as 30 must-win accounts, customer development works.

He argues that a lab experiment is not a proxy for a real world where users are on the subway or raising kids. He says hearing eight customers pontificate for an hour locks things into the room's heads, even when they know the sample is unrepresentative. He suggests reading Shakespeare for insight into humans, while remembering that, like a movie, you must sell tickets and fix the trailer, casting or script when you do not.

He uses DoorDash and Airbnb as examples. DoorDash's founders had a statistic that 93% of US restaurants did not deliver, and Rabois recalls thinking of Andrew Mason's line that phones need an "I'm hungry" button. Airbnb's Brian Chesky showed that roughly 30 Craigslist posts in the Bay Area asked to rent a bedroom. He says random samples would likely have said no to both.

Rabois links this to the ugly baby idea from Ed Catmull's book and says Taylor Swift's advice to iterate privately for a long time makes a similar point. Artists who start taking feedback after a first hit can produce derivative work. As an investor, he wants half his VC friends to laugh at a seed or Series A investment, because ugly babies are where real alpha lies. Lenny adds that early employees at OpenAI and Palantir described the same pattern.

To judge an AI startup, Rabois asks first whether foundation labs will become so proficient that there is no oxygen left, since a company has to be durable for decades. He then asks what its accumulating advantages are, which over time make the business better and easier. Network effects are one example, and at seed stage he wants founders to articulate conceptually where they could build such advantages and when they would start measuring them.

He describes himself as a founder-driven investor: if a founder has a non-zero chance of changing an industry or the world, he is in at seed or Series A. Other investors are technology-driven or product-and-market-driven, and he names Marc Andreessen, David Weiden and Alfred Lin as examples. His question is whether this founder could be the next Brian Chesky.

Among thriving companies he sees three signals. The first is speed: Roelof Botha told him that issues identified at one board meeting were shipped and solved by the next, and Rabois gave Ramp a Series A term sheet partly because it was about three months from shipping a card, which usually takes nine to twelve. The second is talent density, and the third is skipping senior external hires in favor of promoting from within, as at Ramp and Trade Republic.

On hiring, he distinguishes value creation from value preservation, saying experience helps with preservation but less with creation. He gives chief-of-staff roles as an example of an internal talent factory, noting that one CMO and a new head of product both came through that path. Lenny points to Ramp's days counter, which Rabois says opens every board meeting.

Rabois says he picked up the practice of criticizing people in public from a great founder he works with. His logic is that private negative feedback optimizes the individual rather than the system. In public, colleagues see the issue is being handled, others can offer help, and it becomes team-building instead of leaving the rest of the company worried that a problem persists.

He allows that there is an art to it and that good sports coaches mix public and private feedback. Asked about psychological safety, he rejects it, saying high-performance machines do not have it and are about winning. He recommends *The Jordan Rules* and *The Last Dance*, and says that if you want to be Michael Jordan you have to act like him.

In failure corner, Rabois mentions being acquired into Google for around $187 million, well short of his ambitions. He notes that early-stage investors with a 30 to 40 percent hit rate are doing great, so failure is routine. He declined a board member's suggested retros on failures at a thriving company, because he does not want to deter ambitious shots on goal.

Rabois's top book recommendation is *The Upside of Stress* by Kelly McGonigal, which he says argues that to be happy, healthy or wealthy you need more stress, not less. His recent TV pick is *Nuremberg*, which he calls useful for understanding historical tragedies and how to prevent them. He also credits Eight Sleep, in which he is an investor, with transforming how he sleeps.

His life motto is "no days off," which he says comes from Bill Belichick's chant. He says it is a proxy for not making excuses and says he has missed seven days of workouts in seven years. Asked about PayPal alumni who over-performed, he says he had a good sense for who had founder-level ambition, so none surprised him.