Ravi Gupta - AI or Die - [Invest Like the Best, EP.411]
What this covers
Patrick hosts Ravi Gupta, a Sequoia Capital partner and former Instacart CFO, to talk through his blog post "AI or Die." The conversation works partly as an operating manual. Gupta urges leaders to go role by role and ask what each job does for a customer, and whether AI could replace or augment it today and in six months. He frames this as Enthusiastic Re-Underwriting, built on the rehire question, and pairs it with Convenient Belief Examination. The belief he singles out is that a company can transition gradually. He also describes a "shared delusion" in which operators present boards with 20 to 30 percent reductions in a sliver of the business.
The range is wide, and several positions are flagged as contested. Gupta argues that private companies have an edge because public leaders feel bound by what they told investors. He cites Satya Nadella's move from the Enterprise Metaverse to a $100 billion CapEx and AI push. He draws on Dario Amodei's picture of a country of geniuses in a data center to argue companies could rebuild their output with far fewer people. He also predicts the end of seat-based software pricing, with buyers paying only for completed work. Other threads include building products that don't quite work yet, the role of Shared Context in tools like Cursor, and treating a model as a brilliant new hire worth fighting to make work.
The episode closes on two side tracks. One is investing by Slope vs Intercept, which means paying up for steeper growth but requires being early. The other is education: with future tools unknown, Gupta would teach children behaviors such as ambition, curiosity, resilience, and high agency.
Gupta argues that AI is removing the constraints that once limited small teams, so companies that cling to headcount, process, and past public commitments will lose ground to smaller organizations that reorganize around customer value and rapid adaptation.
- Small teams with AI can match what large organizations do, and the hidden costs of headcount (coordination, hiring, layoffs, leadership attention) now outweigh their benefits
- Winning leaders are world-class reactors who respond quickly to new information rather than world-class predictors, and they avoid convenient beliefs that protect their past selves
- Constraints that remain are physical rather than digital, and customers will soon expect magic from every product
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Believing you can transition gradually is a convenient belief that feels easy, and anytime a convenient belief concerns something important it should be examined, because it usually avoids the change.
“anytime you have a convenient belief on something important, you should really examine it.”
Investors should focus on the growth slope of a company rather than its starting valuation intercept, which implies paying more for the best companies that have steeper slopes, but requires being early.
“We invest in slope, not intercept.”
Being private is a major advantage in the AI transition because private leaders do not feel bound by what they told public market investors, while many public leaders feel obligated to their past selves.
“being private is a huge advantage is a general matter because you really have to be courageous to not let something that you've said in the past to a public market investor.”
Satya Nadella's Q4 2022 emphasis on the Enterprise Metaverse was replaced within two years by a $100 billion CapEx and AI focus, showing that leaders should ignore what they previously told investors and focus on what is coming.
“Satya Nadella in Q4 2022 is talking about the Enterprise Metaverse.”
Startups can build distribution faster than ever because AI-powered products feel magical, but incumbents with existing distribution that deploy AI are even harder to compete with, and their main constraint is organizational size.
“if somebody has distribution already and they thoughtfully deploy AI and they actually go and disrupt themselves”
Dario Amodei's vision of powerful AI as a country of geniuses in a data center, available to anyone, implies companies could rebuild current output with far fewer people.
“you should be able to recreate what we have now with dramatically fewer people working on it, because there are a country of geniuses available in a data center.”
Most companies are probably far too big relative to what they need to be as public companies, and AI gives them a real reason to become lean and customer oriented.
“most companies are probably way too big relative to what they need to be public companies.”
Following Paul Graham, frontier builders should build products that do not quite work and are too expensive today, because models will improve and get much cheaper, perhaps by a factor of a hundred, within three to six months.
“you should be building things that don't quite work right now with the models and are way too expensive.”
Seat-based software pricing will not be the future, and buyers will pay only for work completed, which could cut revenue by about 60 percent in the short run but create the potential for five times more revenue later.
“And maybe it means you'll have 60% less revenue now if you move to that. But in the future you have the potential for 5x more revenue”
Gupta argues that the easiest way to see which companies will fail to adapt is to notice those that make observers uneasy, and he warns that the belief that AI is not that good is dangerous and often driven by anecdote.
“the anecdotal experience of AI not changing your life is really dangerous right now”
In a fast-changing environment the advantage belongs to world-class reactors who respond quickly to new information, because predicting the future is beyond most people's control while reacting is a choice that individuals and teams can make.
“I think the thing that I do have in my control is the ability to respond quickly to new information.”
Where headcount once signaled status, the new status marker is a small team's ratio of magic delivered to people employed, which can be measured as magic per employee rather than only revenue.
“The ratio of magic to people.”
Leaders should enthusiastically re-underwrite their whole business by asking the rehire question, if we could do it again would we hire this person for this job, and by holding nothing sacred about the company's purpose, delivery method, or pricing.
“I think you have to enthusiastically re-underwrite your business.”
Leaders should treat AI models like a brilliant new hire who is worth fighting to make work, rather than like a poorly chosen hire, because a failure then prompts better management instead of dismissal.
“The thought experiment you and I are talking about is what if you just knew the person was brilliant”
Most of what is missing for AI to be magical is context, and products like Cursor and Cognition succeed because they share the full codebase context and improve with use.
“most of what's missing from the models in order to be magic for you is context.”
Gupta argues that companies are held hostage by what they said to teams or on earnings calls, and that desire for consistency with past selves blocks adoption of the biggest technological change in a generation.
“so many companies are held hostage by what they said to their team or what they said on an earnings call”
The lower the share of senior leadership time spent on how to deliver value to customers in the best, fastest, and cheapest way, the greater the opportunity for value creation, and for many leaders this share is vanishingly small.
“What percentage of your day, if you're a CEO or a founder or a leader, is spent on things that your customer cares about? And I think that number is vanishingly and astonishingly small for a lot of people.”
Large headcount imposes hidden costs such as coordinating people, interviewing and hiring, performance management, and layoffs, which consume leadership time that would otherwise go to customers.
“there's a lot of hidden costs to having a lot of people”
The constraints that historically limited small teams are dissolving because AI lets a small team recreate what large organizations build, so companies that fail to embrace rapid change risk irrelevance.
“the constraints that historically limited small teams are dissolving”
Operators often present AI initiatives to boards as a 20 to 30 percent reduction in a sliver of the business, and boards accept them without believing they are sufficient, creating a shared delusion.
“some 20 to 30% reduction in some sliver of your business to be 'Hey, board, don't worry, we're using AI.'”
A company that can serve as another firm's AI strategy, the foot in the future, gains enormous value because its own capability keeps compounding as AI improves.
“Can I be their answer to be the foot in the future?”
Anything that reduces a company's agility, such as large headcount, adherence to quarterly commitments, or unwillingness to change, is now a massive limiter because change is happening so fast.
“anything that reduces your agility is a massive problem.”
Because specific skills cannot be taught for an unknown tool future, Gupta argues children should be taught behaviors such as ambition, curiosity, resilience, adaptability, and high agency, which are also what companies need.
“There's behaviors, there's feelings, there's approaches.”
Leaders should go role by role to determine what each job does for a customer and whether AI could replace or augment it today and in six months.
“going through role by role and figuring out what does that job actually do inside my company for a customer and if I replaced it with a really well prompted AI, could it do it?”
Customers are divinely discontented, as Bezos put it, so they will come to expect magic from every experience, and companies either deliver it or competitors with far fewer people will.
“either you're going to create magic or someone else is going to create magic with a lot fewer people than you.”
Because AI lets companies pass many competitors at once, a company's only real limit is the scale of its ambition, and leaders get to choose which race they are in.
“you are only limited at this moment by the scale of your ambition.”
Leaders facing AI should not adopt a wait-and-see position just because someone else says so, and should spend weeks or a month trying to prove themselves wrong before deciding; the conclusion must be their own.
“you can't outsource your conviction, which means don't do it because somebody tells you.”
Capital Group's Mark Casey praised Home Depot in the early internet era for item-by-item analysis of product price against shipping cost to decide which items suit online sales, a sign it was ready for changed shopping, and Gupta suspects this helped Casey's investment.
“we went through item by item that we sell and we figured out the price of the item relative to the cost of the shipping.”
Founders in the AI era need ambition, curiosity, resilience, imagination, and adaptability, with adaptability and the willingness to change on a dime with new information becoming newly important beyond grit.
“I think the thing that I do believe is newly important. It's always been important. But more important for me is the agility, adaptability”
Beyond legal governance, the only real job of a board is deciding whether it has the right CEO, and in an AI transition directors should judge whether the CEO will embrace change.
“the only real job of a board beyond some very highly legal corporate governance stuff, it's do you have the right CEO?”
Had Instacart had current AI tools, Gupta would have used customer order history to speed checkout, improved in-stock accuracy, looked for AI-driven delivery batching to cut minutes, and built phone ordering for less digital customers.
“I would be trying to make the order use as much context as I could to make their ordering process be as fast as possible”
Leaders should not prioritize margins on AI-first products because their delivery cost will fall rapidly, but should prioritize margins on non-AI-first products, which represent the durable future.
“If something's an AI-first product, you probably shouldn't care that much about the margins of it”
Companies with physical products face slower AI-driven gains than purely digital companies because physical constraints such as traffic and store labor cannot be fixed by software, so the more digital the business the faster magic arrives.
“the more your product is physical, the more the bottleneck for using the AI is the physical world rather than the software world.”
When AI output fails, the fault usually lies with the user's prompting, understanding, and context provision rather than the model's capability.
“if something doesn't work right with the model, it's your fault, not the model's fault.”
Gupta prepared for an unexpected dinner in 20 minutes by prompting a top-tier model with Deep Research, reading the output on the way, and having a productive conversation, which showed real-work use beats trivial summaries.
“And I use Deep Research. And he said, "No. Can you tell me these four or five things?" I told it four minutes of thinking”
It is easier to find 20 excellent people than to run a 10,000-person organization, so a 20-person company can now do what it never could before and move faster than a large one.
“It's so much easier to find 20 awesome people than it is to run a 10,000-person org that's so far away you can do it faster.”
AI is better at discerning hard-to-see patterns than people give it credit for, so given a framework it can rapidly identify acquisition targets, such as companies in adjacent spaces with many customers, that work as tuck-ins or transformational deals.
“It finds patterns in ways that wouldn't be obvious and it finds them ridiculously quickly.”
Dismissing AI because of personal anecdotes that it is not that great, especially as a contrarian stance, is dangerous, and leaders should instead talk with people devoting their lives to the technology.
“the anecdotal experience of AI not changing your life is really dangerous right now”
When Bill McDermott took over ServiceNow he held about 10 customer meetings a day including weekends and met roughly a thousand customers in the first quarter, which Gupta cites as a sign of an externally focused CEO.
“it was some number that effectively equated to 10 per day, 10 customer meetings per day, including weekends.”
The best AI-era companies will become worth much more while mediocre ones will lose value, widening the valuation gap and making investment both exciting and intimidating.
“the best companies are going to be worth more and the mediocre ones are going to be worth less. I just think the gap is going to widen”
Someone is out there working right now to produce the magic a company promises customers, and leaders should face that competitor with joy and a desire to earn the chance to compete, since customers will ultimately choose between them.
“For a CEO or a leader or anything, the ghost is out there and you get a chance to play against them.”
The people who founded the AI labs were world-class predictors who bet early on scaling laws continuing, though Gupta argues that reacting to current capability facts is not prediction and is more achievable.
“If you just believe that the scaling laws continue, you could bet that this was going to happen.”
Choosing to react optimistically to change is hard but controllable, comparable to the discipline of going to the gym, and it requires fortitude and choice rather than a new brain.
“It's hard the way going to the gym is hard. It's not hard the way creating a new mathematical theorem is hard.”
The Collison brothers describe listening with predatory curiosity, meaning seeking to fully understand what someone is saying and why, which Gupta sees as the stance of the most respected people.
“they listen with predatory curiosity.”
The shift toward small teams does not mean people lose jobs but that each person delivers more magic, letting individuals leave large organizations to build smaller things or join smaller ones with more impact and purpose.
“that's not because that means people will lose their jobs. That's not the point. The point is that each person will deliver more magic.”
Shane Battier's poster-driven fear that someone was always working harder became a motivator, and he eventually met that ghost in Kobe Bryant, showing that a competitor working harder is a challenge to embrace.
“Shane met the ghost when he played in the NBA. Kobe Bryant was that ghost.”
Gupta argues that Sierra exemplifies an AI-native culture because its team focuses on customer needs, and that anytime something can be improved with AI it should be done and shared across the entire team rather than treated as a one-off.
“Anytime you figure something out that can be helped with AI, do it. Tell the whole team.”
To keep agility, companies should stop providing guidance, tell employees that some outcomes will be negative before long-term positives, and acknowledge that change is hard to predict.
“And maybe that means you don't provide guidance anymore.”
Following Matt Kohler's view that a venture firm's job is to see the present very clearly rather than predict the future, Gupta concluded that the present had changed more than he thought.
“Our job is not to see the future, it's to see the present very clearly.”
Just as NBA teams rarely stay good without a painful rebuilding period, companies seeking to pivot while keeping their current business should expect a painful transition rather than doing both smoothly.
“It is really hard for an NBA team to be really good for a long time and never bottom out before they get to be good again.”
The most impactful recent conversations about AI displayed humility and curiosity rather than assertion, with people trying to listen and get closer to the technology.
“All of them, they're not assertions as much as they are curiosity.”
Gupta recommends that companies envision customer offerings that are impossible or too expensive today and build them now, because they will become feasible within six to twelve months.
“let's all use this idea of what is barely possible now and too expensive.”
A friend running a 400-person organization spent enormous time on letting go of 20 people, including severance, telling those who stayed, and explaining durability.
“he had to let go of 20 people last year”
Gupta admits that recommending radical audits is easy for him because he does not carry the responsibility of implementing them, suggesting outsiders may underestimate the difficulty for operators.
“it's easy for me to say because I don't have the responsibility of doing this.”
Patrick suggests the shift in status toward small teams has already occurred, and Gupta agrees that it is happening and is positive because individuals can deliver more impact in smaller organizations rather than being cogs in large wheels.
“Nobody likes being a cog in the wheel. Nobody likes feeling if they left, nothing would change.”
Following HubSpot's Dharmesh Shah, SMB now stands for small and mighty businesses, and the small-team advantage may be larger than previously thought.
“SMB doesn't stand for small and medium businesses anymore. It stands for small and mighty businesses.”
Investors outside venture can make money by identifying public companies that embrace AI and by pressing private equity-backed CEOs to get closer to what AI can do.
“figuring out the public companies that are actually going to embrace this.”
Coach Krzyzewski said he was not a world-class predictor but a world-class reactor after college basketball players began leaving after one year, because he could not have predicted the rule change but played the new game well.
“I am not a world-class predictor, but I am a world-class reactor.”
Gupta cites Shopify's practice of randomly ending recurring meetings as a model for auditing a leadership team's calendar, to find what percentage of company time moves the needle for customers.
“This thing that Shopify does. Shopify will just kill meetings, end recurring meetings randomly.”
Anything that limits a company's ability to adapt is now a hundred times more painful than before, so companies will increasingly appear to have sub-par costs relative to their competitors.
“anything that reduces your agility in any way, anything that prevents you from adapting or changing, is a hundred x more painful than it's ever been before.”
Patrick notes Tyler Cowen's view that most of his writing is now for AIs, generating context that will make them better for him, and suggests more tools should make it easy for optimists to create such context.
“I love Tyler Cowen's idea that most of his writing now is for AIs, not for humans”
A Sierra team member showed Gupta that a newer model could be asked whether its instructions were clear, and the model replied by asking for clarification on five or six points, which Gupta found a striking shift in how models can be used.
“And the model came back and said, "No, actually, in fact, can you please clarify these five or six things?"”
Shared context with a spouse lets partners anticipate each other's reactions, and AI collaborators can achieve similar anticipation if given enough context.
“The relationship most of us have the most shared context in is the one with our spouse.”
Every company should not rest until its own product produces an AI aha moment that wows its team in the way that products like ChatGPT and Midjourney wowed users.
“we should have a moment that wows us in our own product in the same way we've been wowed by some of these incredible original products.”
Tyler Cowen writes most of his content now for AIs rather than humans to generate context that will improve future model outputs for him.
“I love Tyler Cowen's idea that most of his writing now is for AIs, not for humans”
A friend at a frontier model company switched to another lab because the pace of progress had changed so much in three months that the difference justified the move.
“The pace of progress has changed so much in the last three months that I had to go to this one.”
Most people's everyday AI interactions, such as Apple Intelligence text summaries, are laughably poor, so the average experience understates what AI can do for real work.
“It's more used for comedy than it is used for actual value right now.”
Startups must be very clear about who they are competing against, because incumbents differ widely in how quickly they use their distribution advantage.
“if you're a startup, you better be really clear about who you're going up against.”
The only effective way to lead through AI change is with optimism, because the title AI or Die sounds negative but the message is that companies can make magic for customers with the new tools.
“I think the only way to really lead through this is to lead through optimism.”
Doug Leone repeatedly framed the world as one of accelerating change, which Gupta adopts to frame board questions about CEO capability.
“In a world of accelerating change, this is something Doug Leone would always say.”
Seeing excellence at the highest level gives leaders a bar to compare against, which is a competitive advantage in decisions and hiring.
“if you just see excellence in people at the highest level, it's a huge competitive advantage”
Gupta says public company board members have limited influence, so their most valuable contribution is asking a good question at the right time, and the current question should be how the board feels about the CEO's ability to lead through AI.
“the one or two ways is really asking a good question every once in a while that makes a big difference.”
Gupta argues that AI systems will not replace collaboration with experienced colleagues until they have accumulated the same shared context and can anticipate what users want, which requires deliberate context-building by users.
“There are people you've worked with for a long time where they just know what you're going to want.”
For early-stage investors who pick the right founders building the right things, the upside has never been better, and the key is picking the right people and being chosen by them.
“the upside is better than it's ever been. That I think is an amazing place to sit and it's about picking the right people”
Sam Altman's stated expectation that in a decade every person will be more capable than any person is today is, taken literally, mind-bending and sets the scale of expected change.
“In a decade, every person will be more capable than any person is today.”
Gupta argues that the $200 per month price of the top model tier is not a lot of money when compared with what it offers, which he describes as seeing the future, so cost objections to premium AI are misplaced.
“sometimes people just don't want to do that because $200 a month is a lot of money. It is, but it's not a lot of money when you think about relative it to seeing the future.”
Gupta reports that using Deep Research for four minutes before a dinner gave him a usable understanding of a company's business, which led the board member to think he was well-informed.
“I told it four minutes of thinking and it came back and I read it on the way to the dinner and we had a great dinner.”
A $10 billion company could become a $100 billion company and a $3 trillion company could become $30 trillion, because AI expands what leaders can attempt.
“So if your company's worth $10 billion, this is a chance for you to be worth a hundred billion dollars.”