Max Levchin, Founder & CEO @ Affirm:The Biggest Surprise Scaling to $18.7BN Market Cap
Summary
Levchin rejects treating A/B labels as knowable at hiring: interviews can screen out C and D performance, but the distinction between a good hire and a brilliant one emerges only after joining. Strong opinions and extreme personalities are often worth their quirks; the danger is insecure B players who become managers and hire weaker teams to become “the tallest mushroom.” His non-negotiable is integrity: “Once I can’t trust you, I can’t ever trust you.”
Founders cannot expect employees to inherit their obsession, because the founder’s incentives, passion, and goals are fundamentally different. The leadership task is to discover what each person wants—quality, mastery, or a personal goal—and connect it to the work. Someone motivated by craftsmanship needs help making “every pixel” exceptional, not another demand to work Saturday and Sunday.
Affirm’s postmortem system converts operating failures into an internal library of institutional knowledge. Every blip gets a segregated data stream, a directly responsible author, a written analysis within at most two weeks, and senior review that asks whether it explains the cause rather than merely describing the event. The process must be clinical: “Get rid of all emotion; exactly describe what happened.”
The founding bet behind Affirm was deliberately uncalculated: remove fine print, never build economics around confusing customers, and accept the possibility that the model might fail. Affirm is now public and financially successful. A later infrastructure failure illustrates what “do the calculating” means: although Affirm scaled both front- and back-end infrastructure for a projected 10x launch, the front tier outpaced the back tier, creating too many backend connections and leaving the database in shambles when actual demand was modest.
Speed matters, but an MVP becomes garbage software when a company mistakes evidence of demand for permission to leave the prototype in production. Affirm lets a prototyping team ship rough experiments, then pulls successful features down and rebuilds them for beauty, scalability, and durability. Levchin reserves extra time for “the quality of the invisible parts,” particularly underwriting, infrastructure, and decisions that will persist for years.
Hybrid work is an output problem, not a location ideology, but “enough together time” is non-negotiable. Spending hours on Zoom after an unnecessary commute adds little; never meeting teammates sacrifices trust, shared context, and camaraderie. The same emphasis on presence shaped Levchin’s layoff lesson: leaders must own the responsibility and “be part of the goodbye,” not retreat to their offices.
Levchin said Klarna announced 32% growth last quarter, while Affirm announced 35% growth over the last few quarters; he said Affirm had been taking share based on estimates from those periods. He attributes that partly to serving terms from 45 days through four years and avoids speculating about competitors’ corporate strategies, preferring product and sales teardowns because “someone is making mistakes for us.” As a public CEO, he focuses on Graham’s long-term “weighing machine,” even removing Affirm’s ticker from his desktop to avoid emotional reactions.
An AI narrative without operating substance feeds the market’s voting machine rather than the business. Levchin agrees pure LLM efficiency may be approaching an upper limit, but expects systems built around LLMs—reasoning loops or multiple models debating—to amplify capability. His standard is simply whether the company is using “the most interesting available tools,” not whether it can manufacture a fashionable story.
Deep dive
1. Extreme talent pays only while ego stays humble and integrity remains intact
Levchin’s starting point is pragmatic: brilliant people often come with “extreme personalities,” and most are worth the quirks. Managing them means judging how extreme is too extreme, what a behavioral tailspin could cost, and whether the manager can still bring out their best work.
Opinionatedness is “exceptionally valuable” when a teammate can say the work is “a pile of garbage” while preserving respect for the people who made it. The ideal combines an unsoftened verdict—“this is not going to ship”—with collective responsibility to regroup and produce something better.
His feedback test distinguishes constrained performance from limited ability. If circumstances prevented someone’s best work, tell them the result is embarrassing but their capability is not; if they genuinely cannot do better, directness is secondary because the impending conversation is, “I don’t think you have a place on this team anymore.”
Narcissism turns useful intensity toxic because criticism becomes proof of the critic’s superiority. Humility keeps the work collective; integrity is the harder boundary: “Once I can’t trust you, I can’t ever trust you.” Situations can turn toxic and pass, but Levchin believes people change far less.
2. Incentives beat founder-style exhortation—and letter grades obscure talent
The host’s pushback is recognizable: in sales there is always another gift, thoughtful comment, or resource to send, yet employees rarely share a founder’s “I will do everything to win” fixation. Levchin’s answer is incentives—the founder owns the company, project, and passion, so motivation arrives pre-installed.
Drive and grit are rare and difficult to teach, but no entrepreneur succeeds alone. Telling someone to surrender every weekend because the founder wants success will not create alignment; leaders must identify whether that person seeks quality, mastery, a particular personal goal, or a lifestyle that is incompatible with the role.
Levchin calls the idea of knowing whether every hire is an A or B player across a 2,000-person company “just silly.” Candidates are coached into exceptional interviewees; tests can remove obvious C and D performers, but even coding exercises are rehearsed and gameable. The revelation comes after hiring: “That’s a brilliant hire” or merely “a good hire.”
He accepts the Steve Jobs maxim that A players hire A players, but supplies fear as the mechanism behind its darker side. A players want colleagues who raise their standard; insecure B players fear being exposed and may hire weaker people to become “the tallest mushroom.” Steady B players remain useful, handle work stars may avoid, and can grind their way into A performance.
3. Calculated risk begins by admitting what cannot be calculated
Affirm’s original thesis met bankers who argued that confusing customers was not merely one industry model but “the only way” to make money. Levchin had no Excel proof for eliminating fine print; he simply preferred testing “never screw the customer, let’s see if we can make some money” over building another company inside the prevailing mold. Affirm is now public and financially successful.
Pressed for a well-calculated decision that still failed, he first gives the defensible answer: proper calculation includes hedges, so even a miss should remain survivable. The host rejects that as the “PR answer,” prompting Levchin’s broader concession: many decisions amount to staring into the void and honestly saying, “I have no idea.”
His concrete failure came before a launch forecast to reach 10x existing transaction volume. Affirm scaled both front-end and back-end infrastructure at breakneck pace, yet the launch itself was modest. The front tier had been scaled faster than the back tier, creating an oversubscription of frontend connections that overloaded the database and left the data layer in shambles for several hours.
The lesson was not that uncertainty made the incident unavoidable. Someone should have modeled the architectural oversubscription: a 10x mismatch between front-end connections and back-end capacity would choke the database. “Do the calculating” means finding that foreseeable failure path before the hedge becomes the hazard.
4. Postmortems need segregated evidence, named owners, and no theater
Every operational “blip” at Affirm receives its own Slack thread, document, or other segregated evidence stream. Mixing simultaneous incidents makes causal reconstruction harder; plentiful raw data is desirable, but it must be preserved separately so unrelated log events do not become a false explanation.
A named author then distills pages of evidence into something closer to a white paper. The key cultural constraint is removing apology, self-protection, and ritual promises to improve: people directly responsible must be able to say they screwed up without recasting events to protect their reputations.
The process has enforcement, not just aspiration: a directly responsible individual, managerial accountability, a normal deadline of one week and at most two, followed by weekly review. Senior operators decide whether the document genuinely explains the incident or merely narrates it—one reason Levchin wants experienced A players in the room.
He learns more from failure because successful funnels blur causality: several stages converting at 95% all look good, obscuring where 96% was possible. Even after victory, his preferred question is “what happened to the other 2%?”—perhaps poor parenting, he jokes, but “a great way to run a company.”
5. One-way doors hide inside architecture, contracts, and reputation
The one-way/two-way-door discussion distinguishes public controversies that can fade from the news cycle from decisions that remain durable. Levchin says rational people retain long memories despite rankings, charts, and reputation management.
Truly irreversible choices alter important relationships or reputation, lock in an uneconomic contract, or consume a year rebuilding architecture until sunk effort and organizational disruption make reversal impractical. Postmortems should identify decisions mistaken for two-way doors and record why the organization could not simply walk back through them.
His own calibration runs the other way: for every ten decisions he initially considers irreversible, perhaps only one or two prove so. The useful operating rule is therefore neither “everything is reversible” nor paralysis—it is to distinguish durable constraints from choices whose apparent permanence is mostly psychological.
6. Writing preserves institutional memory; brevity makes it usable
Affirm’s postmortem repository reaches back to the company’s beginning, turning outages and process failures into internal case studies rather than oral tradition. Future teams can retrieve exactly how the front end once overloaded the back end instead of relying on whoever happens to remember it.
A writing culture still needs quality control, and Levchin’s governing standard is simplicity. A feature confined to roughly 25-by-25 pixels may warrant two or three pages, not twelve pages connecting it to grand social consequences; long prose increases the odds that the operative detail disappears.
Live rants can motivate or align teams around a strategy, but at an all-hands a speaker may say 12 things and employees remember two. Notifications and limited attention make the same problem worse in documents, so “the shorter the better” applies in speech as well as writing.
7. Prototype quickly, then rebuild what earns permanence
“Mind the quality of the invisible parts” comes from the story of signing the inside of an early Mac: work should be beautiful enough that its builders would proudly sign components customers never see. That ideal creates a healthy tension between rapid shipping and designing for the next version.
Speed is one of the most important determinants of startup success, although some leaders look composed above the water while their feet move frantically below it. Levchin says his own pace is visible “from the feet up.”
Refactoring and technical debt are inevitable in a decade-old startup, but underwriting models, scalability, infrastructure, and other “crown jewel” systems cannot be allowed to remain merely adequate. Levchin dislikes hearing that work slipped a month, yet accepts delay when the component will matter for months or years.
Affirm’s prototyping team is empowered to ship rough, non-production-quality ideas to customers for evidence. Once data shows traction, the company pulls the experiment down and rebuilds it properly; leaving it untouched produces “a collection of garbage,” while someone else may build the durable version first.
8. Presence matters most during collaboration and organizational grief
Affirm’s offices range from perhaps half-full in San Francisco to standing-room-only in New York. Levchin values output over physical location: focused people may produce more from a spare bedroom, while commuting merely to spend the day on Zoom can be an indefensible use of time.
His hard boundary is shared experience. Teams must meet, learn one another’s decision styles, break bread, and form camaraderie; claiming equal effectiveness after never meeting colleagues is “a stupid argument.” He also says sitting a C-level team together among A players can push its members toward better performance.
Layoffs taught the same lesson about presence. Early in his career, Levchin wanted to hide from responsibility until an executive told him to help people pack and join the grieving floor. “You cannot hide from the grief.” When Affirm later cut staff after growing and hiring too fast, employees told him they understood what had happened, loved the company, and hoped to return—evidence that a strong culture can soften the blow.
9. Market share, regulation, and AI stories all face the weighing machine
Asked whether Klarna was growing faster, Levchin said Klarna announced 32% growth last quarter, while Affirm announced 35% growth over the last few quarters. Klarna’s numbers were not yet public, though Levchin said it was about to go public; based on estimates from the last several quarters, he said Affirm had been taking share against competitors.
His product explanation is breadth: financing from 45-day terms through four years, whereas many rivals specialize in a narrower use case. He declines to analyze Klarna’s U.S. strategy without its information. CEOs should study competitor products, sales efforts, and partnerships—“someone is making mistakes for us”—but divining another company’s master plan resembles forecasting weather 12 weeks out.
On regulation, Levchin favors thoughtful rules such as a late-fee cap and says companies should not build profit models around late fees. He warns that simplistic, black-and-white rules can have a chilling effect and make useful products nonviable, although he says regulation has historically been mostly positive. Companies should engage directly with regulators and explain which proposals are good or bad.
Public markets provide capital and access to sophisticated investors, but Levchin tries to starve short-term emotion. He removed Affirm’s ticker from his desktop, warned employees not to anchor on post-IPO prices, and prioritizes profitability, revenue growth, and net-income growth—the fundamentals measured by the long-term “weighing machine” line attributed to Warren Buffett but identified by Levchin as coming from Benjamin Graham.
The same standard governs AI. Levchin agrees pure LLM efficiency may be nearing an upper limit, yet expects reasoning loops and competing models to extend the system. A public company needs substance, not an “AI story”: use the best tools available, manage growth and profitability, and spend less time feeding the voting machine.