Summary

Ravi Gupta’s “AI or Die” call is that agility has become the scarce corporate asset: if “a country of geniuses in a data center” is even directionally right, today’s work can be rebuilt with dramatically fewer people. Large headcounts, earnings commitments, and promises to a team can trap incumbents in their past; his vivid threat model is a Friday-night crew ranking companies by market cap, employee count, and NPS, then racing to rebuild the largest, least-loved target.
The practical mandate is to “enthusiastically re-underwrite” the entire business from the customer backward, holding nothing sacred—not roles, meetings, pricing, margins, or guidance. Ask whether AI can replace or augment every role today and in six months; seat pricing might give way to paying per completed job, perhaps cutting revenue 60% now but opening 5× more later. Products that barely work and cost too much today may be the right bets because models should improve while costs fall dramatically.
“Small and mighty” teams could turn headcount from a status symbol into a liability, with “magic per employee” becoming the more revealing measure. A 400-person organization’s layoff of roughly 20 consumed an extraordinary amount of time; coordination, hiring, PIPs, and internal reassurance all subtract from customer work. AI’s promise is not merely fewer jobs, but more impact and purpose per person.
Leaders can underestimate AI when they test it like a weak consumer app rather than a brilliant new hire that needs context and management. Ravi’s $200-a-month model converted 20 minutes of preparation into a strong dinner briefing after four minutes of deep research, while Apple Intelligence text summaries represent the laughably weak experience many people know. His deliberately severe rule: when the model disappoints, first examine your prompting, context, and understanding.
The startup-versus-incumbent race has accelerated on both sides: magical products can acquire distribution faster, but incumbents that pair AI with existing distribution become more formidable. Patrick cites Cursor-like companies reaching $100 million of revenue within a year using 10–30 people; Ravi points to Microsoft as harder to attack because Satya Nadella understands the opportunity and is pushing AI into products with immense distribution. Organizational size—not Satya’s intestinal fortitude—is the constraint on speed.
Investors should expect a wider power law: “the best companies are gonna be worth more and the mediocre ones are gonna be worth less.” Sequoia partner Pat Grady’s “slope, not intercept” framing suggests exceptional trajectories might justify higher prices, but finding such companies earlier matters more. Ambition, imagination, curiosity, adaptability, and the ability to become a “world-class reactor” are especially valuable. Boards should ask whether their CEO inspires confidence amid accelerating change—and free the right one to “play free.”
The optimistic case is that AI lets ambitious, high-agency people enter races previously closed to them—and win with 20 exceptional people rather than a 10,000-person organization. Ravi’s “ghost” is the unseen competitor already building magic, like Kobe Bryant eventually personifying Shane Battier’s childhood fear; his counter-image is Ayrton Senna passing 15 cars, turning a $10 billion company toward $100 billion or a $3 trillion one toward $30 trillion. “What a time to be alive.”

Deep dive

1. The present changed before most leaders noticed

Ravi’s wake-up began with Sierra: Bret and Clay described tasks o1 Pro could perform that 4o could not. Clay then asked the model, “Are my instructions clear?” Instead of proceeding, it requested five or six clarifications—an interaction Ravi had never considered and found “pretty stunning.”
A friend switching model companies supplied the second signal: “The pace of progress has changed so much in the last three months that I had to go to this one.” Matt Cohler’s maxim—“Our job is not to see the future, it’s to see the present very clearly”—made Ravi realize the present itself had outrun his understanding, despite his working in Silicon Valley venture capital.
The essay’s conditional is extreme but explicit. Dario’s October vision offers “a country of geniuses in a data center,” while Sam’s claim is that “in a decade, every person will be more capable than any person is today.” If those claims are even approximately right, every company and individual faces dramatic change.
Ravi’s first response was personal, not commercial: he and Avni asked what their children should learn. They concluded that durable skills were hard to specify, but ambition, curiosity, resilience, adaptability, and high agency would matter—the same behaviors companies now need.

2. Corporate history has become a dangerous sunk cost

Ravi’s sharpest disruption image is a group of developers spending Friday night sorting companies by market cap, employee count, and NPS, then competing to rebuild the largest, most overstaffed, least-loved target. It sounds wild only if powerful AI fails to approach the capabilities its builders predict.
His operating test starts role by role: what does each job actually deliver for a customer, and could a “really, really, really well-prompted AI” replace or augment it today? Repeat the question for six months from now, because bleeding-edge builders should pursue things that barely work and are too expensive before models improve and costs fall toward “a hundredth of the price or whatever.”
Pattern-finding expands the opportunity beyond cost reduction. AI might identify non-obvious acquisitions that improve distribution, while a company could become another business’s AI strategy or answer for the future. Ravi calls that position “absolutely gold” because the vendor can expand into more jobs as the models improve.
Microsoft illustrates the required disregard for prior commitments: Satya Nadella discussed the enterprise metaverse on the Q4 2022 earnings call, then the company’s current discussion became roughly $100 billion of CapEx and an AI-centered agenda. Ravi’s lesson is categorical: treating old guidance, internal promises, or strategies as binding can block “the biggest technological change that we’ve probably seen in our lifetimes.”

3. AI becomes useful when it is treated as a colleague

With 20 minutes before an important dinner, Ravi told the $200-a-month model that he was an investor, unprepared, and needed a deep explanation of a company plus useful ideas for the person he was meeting, who sat on that company’s board. After requesting more context and thinking for four minutes, Deep Research produced a briefing he read en route—and the conversation went well.
That is “real work,” not summarizing an email. Ravi contrasts it with Apple Intelligence’s text summaries, which Patrick calls “terrible” and Ravi says are used more for comedy than value; weak consumer experiences can therefore create dangerously weak intuitions about what the technology can do.
Patrick and Ravi’s thought experiment is to treat the model as a brilliant recruit who will work 24 hours a day. A flawed first project would prompt better management, not immediate dismissal. Instead, people treat AI like “someone’s nephew that we had to hire,” then declare that it fails.
Ravi’s rule is intentionally demanding: “If something doesn’t work right with the model, it’s your fault, not the model’s.” Most missing magic is missing context. Cursor and Cognition benefit from understanding the codebase; spouses can communicate with an eye flick because of shared history. Patrick adds Tyler Cowen’s idea that writing now creates context for future AIs.

4. Headcount hides costs that customers never value

Dharmesh Shah’s revised acronym captures the “small team meme”: SMB now means “small and mighty businesses.” The emerging status marker may not be how many people a leader manages, but how much magic a small group produces.
Salary understates an employee’s total cost. More people create coordination, interviewing, hiring, PIPs, and internal communication; one friend running a 400-person organization spent an extraordinary amount of time designing a roughly 20-person layoff, determining severance, speaking with departing employees, and reassuring everyone who remained.
Ravi’s proposed diagnostic is the percentage of leadership time spent delivering customer value in the “best, fastest, and cheapest way” over six, 12, 18, and 24 months. For many CEOs that share is “vanishingly and astonishingly small,” explaining why some dislike jobs dominated by internal machinery rather than customers.
Bill McDermott provides the external-focus exemplar: Ravi recalls his claim of meeting 1,000 ServiceNow customers in his first quarter—roughly 10 per day, including weekends. Sierra similarly keeps asking what customers want and how fully AI can deliver it. Being private can make this flexibility easier, though Ravi says Fidji will do what is right for Instacart regardless of whether it is private or public.

5. Magical products compress the distribution clock

Patrick updates the classic contest—whether startups gain distribution before incumbents gain innovation—with companies such as Cursor going from zero to $100 million of revenue in a year with 10–30 people. Distribution itself now appears to arrive at unprecedented speed.
Ravi thinks the old question still holds, but the clock has accelerated: customers adopt products that “feel like magic,” and AI can supply that sensation. He hedges that he has not studied the history enough to know for certain, but his guess is that a well-executing startup can now distribute a magical product faster than ever.
The inversion cuts both ways. An incumbent that thoughtfully disrupts itself can combine magic with enormous installed distribution; Microsoft is therefore “way harder to compete with” than before. Satya’s intestinal fortitude is not the constraint—organizational size is. Startups must distinguish such rivals from large companies whose internal friction prevents equally fast deployment.

6. Becoming AI-native requires changing the economic model

Ravi’s prescription is to “enthusiastically re-underwrite” the business: restate its purpose for customers, identify the best, fastest, cheapest delivery method, and hold nothing sacred. Seat-based software pricing may give way to payment for completed work; a transition might mean 60% less revenue now but potentially 5× more later through more customers or jobs.
The audit then moves through every role and calendar. Determine the “AI superpower version” of each job—replacement or augmentation—and ask what percentage of time engineers spend coding, alongside measuring how much leadership and company time moves customer outcomes. Shopify’s willingness to kill recurring meetings offers the model: internal routines do not earn permanence merely by recurring.
Margin discipline should split by product type. Ravi would tolerate poor margins on an AI-first offering because delivery should become much cheaper, but scrutinize margins on non-AI products as their economics are more durable. He would also build what the company has dreamed of providing customers but previously found impossible—especially products that are barely feasible and uneconomic today.
Patrick supplies the strongest pushback: a battleship CEO might fear wasting two years on another metaverse-style wild goose chase. Ravi concedes that advice is easy from outside but insists, “You can’t outsource your conviction.” Spend weeks or a month actively trying to break the convenient wait-and-see belief; if it survives, defer affirmatively rather than by inertia.

7. Reacting well matters more than forecasting perfectly

Coach K’s line gives Ravi the attainable standard: “I am not a world-class predictor, but I am a world-class reactor.” He could not predict college basketball’s shift from four-year players to one-and-done talent, but once those were the rules, he played the new game exceptionally well.
Ravi finds that framing liberating because reacting requires no prophetic IQ. Leaders can seek new information, resist aversion to change, and respond quickly. It is “hard the way going to the gym is hard,” requiring discipline and fortitude, rather than hard like inventing a new mathematical theorem.
Instacart makes the customer-backward exercise concrete: ingest a shopper’s previous order—even from elsewhere—to shorten checkout, improve store-level availability so unavailable items never appear, optimize batching to cut delivery minutes, and perhaps make phone ordering economical for people, possibly older customers, who are less familiar with the app. But Ravi preserves the physical constraint: AI does not fix traffic or make shoppers run faster.
Digital-only businesses should therefore feel the impact sooner than physical ones. Yet expectations spread across both: Bezos’s “divinely discontent” customers will demand magic everywhere. “Either you’re gonna create magic, or someone else is gonna create magic with a lot fewer people than you.”

8. The investment gap widens around adaptable leaders

Pat Grady’s formulation remains Ravi’s underwriting anchor: “We invest in slope, not intercept.” AI can make the best slopes much steeper, potentially justifying higher valuations—but it also raises the premium on finding those companies earlier and identifying founders with ambition, imagination, curiosity, resilience, optimism, and extraordinary adaptability.
The newly weighted trait is the “world-class reactor”: a founder willing to change on a dime while preserving only the obligation to deliver customer magic. Ravi’s conclusion is both opportunity and warning: “The best companies are gonna be worth more and the mediocre ones are gonna be worth less,” so investors’ selection errors become costlier.
Patrick asks whether investors outside Silicon Valley could exploit slower peers through public equities or private equity. Ravi says “I don’t think it’s better,” only that there are multiple ways to express the thesis. Find public companies genuinely embracing AI or help private-equity CEOs engage deeply—but be wary of board slides touting a 20–30% reduction in some tiny business sliver as proof of transformation.
For boards, beyond highly legal corporate-governance duties, the decisive question remains whether the company has the right CEO. In accelerating change, the answer should feel either “amazing” or unnerving. If the CEO understands the opportunity but fears board reaction, directors should say, “We want you to play free”; if capability or appetite is missing, they must confront it.

9. Small-team status turns competitive fear into ambition

Ravi sees “magic per employee” as a possible replacement for team size as status: “Oh my God, you did that with ten people?” The aim is not job loss for its own sake, but people leaving cog-like roles for smaller teams where they create more, matter more, and feel greater purpose. Many companies, he suspects, were already bloated before AI supplied a reason to change.
The most credible recent conversations carry “insane humility,” not assertions: “Wow, there is a lot happening. I don’t totally understand it.” Ravi prizes people seeking proximity to the technology and truth with what the Collisons were said to call “predatory curiosity”—a need to understand what someone means and why.
Shane Battier’s childhood “ghost” was the unknown player training while he rested; in the NBA, he met him in Kobe Bryant, born one month later and raised in Italy. Kobe became the comparison for the ghost that had pushed Shane through his work and into a 13-year NBA career. Every CEO now has a similar ghost building the magic promised to their customers. The right response is joyful competition—but “the ghost is good.”
Ayrton Senna supplies the optimistic closing frame: disruption is the moment to pass 15 cars, and each company chooses its race. A $10 billion business can pursue $100 billion; a $3 trillion one can pursue $30 trillion; a new 20-person company can challenge organizations of 10,000. High-agency people have tools to “enter a race you never knew possible and to fucking win it.”

Why is everyone cloning Deep Research?February 18, 2025