Posting a photo of this moment was not a product category yet

Lex Fridman opens with a greatest-hits setup: Massachusetts, Doom II, a vinyl shop, Stanford, turning down Mark Zuckerberg, a semester in Florence studying photography. Kevin Systrom, co-founder and longtime CEO of Instagram, answers with the unglamorous origin. In 2010 the hot product was the check-in. Gowalla, Foursquare, and Hot Potato let you tell the world you had arrived at a place. Systrom sat down to build a better Foursquare, added extras including a photo of where you were, and recruited Mike Krieger. Nobody used it. "No one cared," he says, "because it wasn't different enough."

The pair had a come-to-Jesus afternoon. If they were going to do a startup, they could not keep shipping the current thing. They wrote down three features people uniquely loved in Burbn, spelled B-U-R-B-N, and photos sat at the top. Facebook at the time wanted an album of your vacation from two weeks ago. Twitter let you attach a picture, but the feed was mostly text and did not show the image inline. "Posting a photo of what you were doing at the moment was like not a thing." Check-in apps had you declare a location and optionally add a photo. Instagram reversed the order: the photo *was* the check-in. Systrom still thinks they replaced those apps. The ledger just never recorded it that way.

Years of studying machine learning gave Systrom a vocabulary for what they had been doing by instinct. Put the product in front of a hundred people, look at the data, notice that they resonate with Y when you expected X, and nudge the company toward Y. Do that fast enough and you can reach product-market fit. Most companies die on the learning rate: too low, and they sit on a thesis the numbers have already killed; too high, and they chase every new idea and never groove. The list of three loves was just the local maxima users were already showing them.

He does ask people. He also does not trust the answers. After a Thanksgiving he had planned in OmniPlan, a Gantt-chart tool he suspects maybe ten people on earth have bought, he asked the table what was good and what was bad. Everyone said everything was great. He knew he had botched dishes. Ray Dalio had been telling him that honest feedback is scarce; friends and family facing a founder who quit a job to build "something wonderful" will not say the hard thing. One class of people will not criticize even when begged. The other class is negative about everything, which is no more parseable.

His rule is to collect the words and then watch the usage. People are defensive about time. They will tell you they were too busy to open your app while posting on Instagram all day. Time is the currency of a life, and there is no way he knows of to invent more of it. If they do not use the product, it is because it is not great. Lex warns that a single metric can obscure the insight, and Systrom agrees with an aviation picture: stare only at airspeed and you will not know if you are climbing, and you will never land the plane. Airspeed of zero still means you are falling out of the sky. Scan a board of instruments. Data does not lie. One gauge will mislead.

The biggest companies, Systrom argues, are founded on enormous technical shifts. The shift before Instagram was a phone that did not suck. The first iPhone looks primitive now and was still another species next to everything else. The first camera that could go toe to toe with a point-and-shoot was the iPhone 4, which arrived the year they launched. Everyone now had a camera in a pocket, and that camera had a network attached. Desktop social networks lived on links you posted while sitting still. The question was what you would share if you were already out in the world. Competitors saw the same opening. Instagram wrote down what sucked about photos: they looked terrible, uploads were slow, and posting to several networks meant opening each app and sending the file again.

The look came from Florence. A photography teacher handed him a Holga, a cheap plastic camera, originally a few dollars, medium-format film, light leaks and blur that a later generation decided was a style. Why not lean into the fact that phone photos sucked and make them suck on purpose, artistically? The first filter, X-Pro 2, was built in a bed-and-breakfast in Todos Santos, Mexico, on a week away from Burbn. He iterated over the RGB array and shifted each channel with a simple function, mapping one color space into another. It took two or three seconds to render. He only later learned to do it on the GPU with OpenGL. Startups, he says, can get away with slow and terrible as long as the outcome is right, then compress the experience over time.

Three seconds survived because the alternatives were worse. Instagram output 512 by 512 pixels, so they walked far fewer pixels than rivals chasing huge images that could take twenty seconds. The trick that felt like lightning was borrowed from Gmail. On that team, entering a username was such a strong signal you would sign in that the account was already loaded into server memory before the password arrived. Systrom asked Krieger to upload the photo in the background while the user stared at a caption box. If they cancelled, delete it. People hit send and watched a progress bar finish instantly. "We were no faster than anyone else at the time." They had hidden the wait in time the user was going to spend anyway. A shell game, and it mattered.

The nastiest pre-launch fight was smaller. Systrom was cheap and still on an iPhone 3G. Krieger had a new iPhone 4 whose feed was buttery. On the 3G, every photo chunked as memory allocated. They sat on that stutter for days. Far too many teams, Systrom says, obsess over the cool computer-science problem of scaling to a billion users when they have a hundred. The loss function is whether the experience feels magical. If that requires something complex, fine. Most of the time the win is sitting there with an unglamorous fix.

Early Instagram also worked as a single-player game. You could filter photos alone, and some people did not realize it was a social network until a friend showed up, at which point they already had a trove to like. Systrom compares an empty social product to a restaurant with great food and nobody in it. Seeding a small group that produced photos you wanted to see, and connecting people who loved a style from around the world, mattered as much as polish. The app was international by default: you did not need English to look at a waterfall. Translations were pretty bad. The images still worked. The client was Objective-C, iPhone only. The backend was Python and Django, before Python was cool for machine learning, plus Postgres, chosen because they thought they were still a check-in app and wanted the geo features, and Redis early. "Not only hipster photo company. Hipster tech company."

That stack lasted through about fifty million users. There was no machine learning at all. The hard part was not Python. It was where to put data arriving too fast, how to stay up, and whether the boxes had enough memory. At launch they had a single computer rented in colo space in Los Angeles, because that is what Systrom remembered from an internship at Odeo, the company that became Twitter: ear protection, cold air, fans, swapping machines. In a Benchmark office someone asked why they did not add more instances on Amazon. He asked what instances were. That night they moved the database to AWS and did something he still apologizes for: the app server stayed on the little LA box, talking to the database over the public internet. It was unusable. Later that night they put the pieces together. If you think you have no chance because you do not know the tools, he says, that is how they started. The question is how fast you notice the mistake and whether you fix it immediately.

His advice on scale is blunt: duct-tape product-market fit first. He cannot name many startups that went from nothing to something overnight and then could not scale; worrying about fifty thousand users before anyone likes the product is the default death. Tests, which feel slow, let a four-person team touch any part of the code without taking the site down. Go slow to go fast. Fit itself is not an absolute number. If a thousand people sign up each week and those cohorts retain, and earlier cohorts even curve up as friends arrive, that is the signal. Seven billion people will mostly never know you exist. The work is how you turn on the spigot. Money depends on the category. Social networks have really only made money one way, ads, so a service amenable to that can wait and hire for it. Enterprise tools should worry from day one. Waving at a maybe-never metaverse, he says, is a bet only something the size of Facebook can finance.

Asked whether anyone can still launch a Facebook, Twitter, or Instagram-scale network, Systrom says absolutely, not because he knows how, but because of the prior. It has happened every three, four, or five years. After Myspace people said it was over; then Facebook. After Facebook and Twitter did images, people said there was no room; then Instagram. Then Snapchat, then TikTok. You do not beat Google or Microsoft at the core business. Instagram won by doing things Twitter would not, images in the feed, and by living purely mobile while Facebook was still mostly desktop. The best place to play, he says, citing a strategy book on moats he cannot name, is where a competitor is structurally unable to pivot.

Snapchat existed because Instagram had been so focused on manicured, permanent versions of the self. People loved the behavior and still wanted to share something from the day without putting it on a profile forever. That opened a market. Instagram's Explore page, meanwhile, already had the habit of sitting there all day. That is effectively TikTok, except it had to be video, and it had to have music. Those details are the hard, unpredictable part of product. His sharper thesis is that the worst thing about social networks is the people, a line that sounds like a joke because that is why you call it social. One job is keeping up with ties. The other is matchmaking entertaining content to people who want it. TikTok barely does the first job at all.

Recommender systems usually run in two stages. Candidate generation pulls a small subset, typically from what friends or influencers shared. Ranking then boosts whatever gets engagement. Friends share what will get likes, which skews emotional and risqué, so the feed feels divisive even before the ranker runs. Why must candidates come from friends? Why not from everything? TikTok's gift is that you can be anyone: no following to buy, no audience to maintain, only something interesting. Systrom thinks that is the direction, less about who manipulates distribution, more about whether the content is good, with each network defining its own value function. Information retrieval talks about queries and documents. A query can be a person, age, gender, hometown, a sketch of interests. A document can be a video. The whole business of recommendation is the distance between them.

Lex wants metrics beyond engagement: whether someone looking back thirty days can say the hundred hours made them a better person. Systrom pulls the curtain on the meetings. Models emit probabilities of click, share, comment, dwell, even downstream follows. Managers then argue how much a comment is worth versus a click, peek at the quarter's goals, turn some heads up and others down, and blend the mix into one scalar to optimize. You think you are solving for "meaningful interactions." Because you are optimizing it, you also buy a pile of side effects. He is careful: he was not in those Facebook rooms. The structure is the point. At a billion people, unintended consequences show up either way.

The old Facebook argument was that more use meant more happiness. Systrom does not buy the map from time to utility. You can do more of gambling without getting happier. Designing a scalar reward for someone's happiness over time is, he has learned, brutally hard, and reinforcement learning makes the gap obvious: reward design is the job, and the current products do not get enough signal from a human being. He is genuinely looking for people who can model utility without a surrogate. "Please if you know someone smart doing this, introduce me." Rethinking algorithms, rewards, and utility is, in his view, an open company-sized opportunity, even if a newcomer starts with too little data to compete on the old objective.

April 2012: Instagram and its staff of thirteen sold to Facebook for a billion dollars. Systrom had sat in Google corporate development, the group that buys companies, and had not closed a single deal, but he had seen prices and process. They were about twenty-seven, exhausted, trying to keep the thing running. He walked into a firm and asked for a five-hundred-million-dollar valuation. "I've never seen so many jaws drop all in unison." A call the next day said the ask had been pretty offensive. Sequoia, Greylock, and others eventually came in around four hundred fifty million pre, fifty million raised. Facebook, after negotiation, doubled the number to a billion. He might have had ten thousand dollars in the bank. YouTube was the last exit he could think of anywhere close. There was no public market in which to mark Instagram to market. In an illiquid room with one serious buyer, you go with your gut.

The calculation he remembers was not the zeros. Could they strap a little company onto a rocket ship that already had talent and distribution, and get the product to a lot of people fast? He had watched Ev Williams sell Blogger to Google before that IPO. They sold before Facebook went public; the stock later sat at seventeen dollars and he thought, what did I just do. Around three hundred twenty, people still called him a dope, just for a different reason. What he would do differently is hire more carefully. Headcount jumped to a hundred, two hundred, three hundred through Facebook's internal-transfer machine. He never recruited or screened those people. Elon Musk sold PayPal for something in the billion range, then used the capital to build actual rocket ships. Selling, Systrom notes, can free a founder who would otherwise stay welded to one company for a very long time. He stayed six more years anyway. The Pope wanted to press the sign-up button. That is a different game than payments.

For six months after the deal, restaurants and sidewalks were full of talk about the sale. He had made a lot of money and could not understand why it did not feel great. They had won the Super Bowl of startups, and then came "we have arrived" syndrome: if we won, what is next? A product manager left early, saying the company had hit its apex and there was nothing left to learn. From 2012 until Systrom left in 2018, he learned more, and took more humility, than in the years before. A billion is cool. There are hundred-billion-dollar companies, and Instagram still had no revenue. Fighting Snapchat was Yankees versus Red Sox, the thing you live for. Everything is hard, including sitting on a beach angry that the wine did not arrive. Choose the game where the pain is interesting. He still does not have the counterfactual in which they said no. Path turned down a large Google offer and faded from memory. Clubhouse, he would not count out. He is proud of what they built, and it is the next shot.

Lex lists photos, one-minute videos, IGTV, Stories, Reels, Live, and asks for the design philosophy behind the Snapchat war. Systrom points to Clayton Christensen's *Competing Against Luck*, a terrible title for jobs-to-be-done theory. Imagine every product in your life as an employee. You are CEO of your life, and you hire products for roles. Instagram's job is sharing your life with others visually, because you feel connected, you get to show off, you feel cared about. Any product that serves that job will do well. Stories serves it more efficiently at scale, when permanence starts to worry you that a college-admissions officer will see a photo. The original feed still has value; together they work. IGTV, he thinks, probably did not perform the job. Shopping is unrelated. It might still work.

Apple, in this frame, is managing your digital life: products that sync, even with silly cords, as a system. Leave the system for something weird and people scratch their heads. Facebook's recurring miss, he argues, is never fully appreciating the job of the main product, then copying a shiny object that has traction and wondering why it does not fit. Facebook worked at Harvard, then Dartmouth, Stanford, and a handful of schools, because real communities wanted a space to coordinate and communicate. Groups and Marketplace are where it still stands most strongly. Groups lets everyone found their own little Stanford. When the app tries to compete with whatever is exciting this year, consumers get confused, even if the bolt-on eventually works. Entrepreneurs get bored and add things. He almost had a heart attack hearing Lex's feature list.

His main thesis about Facebook's position is simple. If the company had spent a decade shipping launches that delighted people, it would be in a totally different world. Apple had you standing in line. Amazon puts the thing at the door in two days. Both have labor and environmental issues; you do not hear a large chorus demanding they pay for their sins in the same way, because they are still producing hits. A hit is not only a financial one. It is a product that brings joy to some fraction of the population. TikTok, visual and not defined by who you follow, now has more sway over the American psyche than Facebook's algorithm, he says, and people love it. Produce, and consumers stay on your side. Stop, and they start picking you apart, right or wrong.

Why is Mark Zuckerberg distrusted, disliked, sometimes hated? Systrom pushes on the premise, then takes the question. Founder-led companies carry a visible story. Zuckerberg is no longer only a person; he is an image of wealth and power, and this moment, a pandemic, a bitter election, wants somewhere to put anger. The founder-CEO of a social network, he says, chose the harder video game. Elon Musk kept a factory open, said offensive things, flipped off the SEC, and people still like him. Jack Dorsey is respected more than loved. The mechanism Systrom trusts is almost too simple: how a product makes someone feel, they ascribe to the founder.

Amazon is utilitarian brown boxes. People find Jeff Bezos impressive and large, not wondrous, even when he does space. Musk is Mars and the kid looking up asking if there is life out there; that hope gets written onto him. Social networks make you angry, or show you a life you wanted, and that mood lands on the person in charge. Early Instagram, as Systrom remembers it, was wonder and beautiful things more than FOMO, though it drifted later. Steve Jobs was not always liked, and then he delivered a product and people fell in love again. The missing skill has a medical name: bedside manner. Jobs could take a boring competitive question and answer it as if you had asked your grandfather the meaning of life. Without that, and with past missteps, trust is hard to earn back. People do not forget the moment they felt a company did not have their back.

He and Dorsey once sat at a desk at Odeo, writing JavaScript for a boring widget. Systrom was an intern who does not think he provided much value. Ev Williams, feeling people were not working hard enough, brought out a hollow statue of Mary, stuffed what felt like a thousand dollars in the base, maybe a hundred, and had the company vote, closed ballot, for who had worked hardest that week. Systrom got two votes and was thrilled. Dorsey took the vast majority. Systrom could not imagine what that new guy would become. He could already see the dedication, and that other people respected it.

Silicon Valley's backlash against hard work drives Systrom nuts. A TikTok he liked said, "I'm taking a break from my mental health to work on my career." He and Krieger lost years off their lives; he has more gray hair. Most people are not successful, and the ones who are did not skate by. Nobody is required to grind. There is also no path to changing the world without a lot of work, and even the people he respects most only nudge the world a little. Musk, interviewed at the Tesla factory, looked visibly shaken and called the pace unhealthy. "To have great outcomes you actually do need to work at like three standard deviations above the mean."

Every choice has consequences, including children. He loves his two kids and says it is extremely hard to work at that intensity and parent well, which is why some people running huge companies delay or deprioritize kids for a while. The contract has to be clear. Joining a startup that is currently sprinting is like joining the Yankees and then skipping batting practice because you wanted the salary. Rest days, from his years as a cyclist, are not laziness; they are how you train longer. Athletes hate them and still take them, usually because a coach makes them. People also hide behind "long-term optimization" when they are simply not working. Bridgewater looks cultish from the outside and, to Systrom, is mostly honesty up front: we record every meeting, we are brutally honest, and if that is not for you, work somewhere else.

As you ascend, nobody wants to tell you that you are crazy. Asking does not fix it; you have to hunt for markers and stay paranoid. One difference he claims from a lot of entrepreneurs is that he was not completely confident they could stay independent and still be great. Internally he tried to say when he had screwed up. Everyone thinks they have to bat a thousand. The best quant funds bat something like 50.001 percent and take a lot of bets. Steve Jobs was wrong about the Newton, about NeXT, about video on the iPod. What matters is how many at-bats you get and whether you are better on average, then whether you can still admit the miss once ego and a circle of compliments grow. He is almost glad his life is relatively normal, with no fifteen Sherpas fetching water. The next thing may be smaller, which means working twice as hard and learning twice as much. Famous tech men who have never built a machine-learning system still like to tell journalists it will kill us all. Programmers are often not the ones writing the op-ed. He would rather be back in a game where he has to fight.

Frances Haugen's files claimed Facebook knew, from its own research, that Instagram harmed teenage girls. Systrom will not pick a single villain. Is the blame on the people who originated the network, including him? On combining it with another network and another set of values? On how it was run after he left? The driver or the car? The devices, or human nature? Any technology has negative side effects. The leadership question is whether you put enormous resources against them, admit true criticism, stay vulnerable about not being perfect, and accept accountability. People generally respect that posture. What did not age well, he says, is Zuckerberg's line that it was crazy to think Facebook had anything to do with an election. If the public does not believe you feel the pain, the words do not matter.

He is wary of the research slice and of the media machine. Psychology is easy to steer with the questions you ask; one internal study is a thin cut of a huge picture. Negative stories about social networks and famous CEOs travel. Couple a whistleblower, a document dump, and a film like *The Social Dilemma*, and the orchestration is almost too clean. What surprised him is the scale of harm a product manager can do to a very large company by releasing meeting notes and studies. His own office had a no-visitors rule because confidential work sat on the walls, and Facebook people hated that, because Mark sat in the aquarium where anyone could watch. Apple and Snapchat lock down for a reason. He thinks this is a seminal moment. The next Facebook-scale startup may grow up closed. That is a loss for internal transparency, even if "nothing to hide" is the slogan, because you can pick documents to show anything.

He does not like the conspiracy that Haugen is a financed operative. Occam's razor is that she believed something was wrong and put herself on the line against a company that does not love enemies. Lex is more unsettled by the human incentives, fame and money extracted from real problems, and by how that feels to people like Yann LeCun, whose researchers want to push reinforcement learning and teach robots to touch and lift. Systrom will not let the organization off. Great people inside a company are not the same as the company. The money that funds that research comes from an enormously profitable machine with trade-offs. You are not completely separate from the system. Look in the mirror, he says, in the Dalio sense, and fix what is there.

Asked where machine learning will matter, Systrom starts with the unglamorous: climate, wasted energy, logistics, a paper on reinforcement learning for buildings, peak load for servers, then cities. Everything Instagram applied ML to got better for users and for the company, even when he did not understand the math as an executive. Too many leaders talk about it the way they once talked about "doing mobile." The next generation will have taken the classes. Reinforcement learning fascinates him because neural nets can now approximate a Q-function, the value of any action in any state, instead of hoping a linear regression works. The field has something like seventeen acronyms and no consensus, unlike the well-worn path of Adam and ReLU. It is noisy and hard to converge, which is why self-driving needs simulators: you cannot crash real people to teach the model that crashes are bad. The leap he wants is the same idea for conversation and feeds, designed in virtual worlds, then connected back to real interfaces. For a long time Facebook ads were basically logistic regression, a published paper, and it made a lot of money. At extreme scale, he says, we are not yet touching what reinforcement learning can do.

Advice for the kids in a garage is not a secret social-network idea. Choose work that would still have been fun if it failed. They started Instagram because they loved photography and social products, not because the ending was written down. Draw three overlapping circles: what you have experience in, what you are excited about even when it is hard, and what the world needs. Without the third, the other two are a small market or a hobby. Be honest about the first by looking at a track record, not a wish; it is not bragging if you can do it, and too many people are delusional. A former engineer wanted to found a company, and Systrom asked whether he wanted to coach an NBA team or play. Confusing passion with a large market is scary. His own sweet spot is the next five years, something he can ship in a year that people will use at scale. He cannot talk about the metaverse with a straight face. That is temperament, not a ranking.

Funding helps when it comes from the right people, including yourself. Matt Cohler at Benchmark was brilliant and aligned; other founders get a noisy board and a fight. Capital is cheap enough that raising money is better understood as hiring someone for the team, with cash as a byproduct, plus healthy pressure from people paid not to lose it. He had lunch with Musk once, asked about angel investing, and got a straight face: why would I do that, I only invest in myself. Novel, and lonely, because then nobody who would lose the money tells you to face reality. Dorsey still brought partners into Square to build a truth-telling room. Systrom likes burning the bridges behind him so that failure is actually failure. Musk's line, relayed from an early Zuckerberg conversation, is that some people put on a show of hard work and are not working hard. Operating, dealing with real humans, is one of the hardest jobs on earth. Parking money in the S&P is not the same game.

Love, in this telling, is Nicole, his wife, who had the filter idea and, more importantly, saw him before Instagram and could tell when he drifted. She said he should do it, and if it failed they could move to a smaller apartment. That permission to fail is almost a superpower, and it somehow leads to success. She is also the least impressed person he knows: the product is fine; she loves him, and she likes that he is a decent cook. Krieger is the other half of the same pattern. Neither of them alone would have made Instagram; they pushed and pulled. Meaning of life: he has not found it. Every level of success still brings the same worries at a different scale, which is how you start believing there is more machinery underneath. He uses *The Truman Show*: chase money and fame until you hit the wall, then notice what actually matters. That is not an argument for a four-day week. It is an argument for designing the life you are in. He is not afraid of dying. He is a pilot and does crazy things. He is afraid of waking up after ten years of a life he never chose. Skip the ten-year plan. Plot a course, and every day, opt in. Fridman closes with Systrom's own line: focusing on one thing and doing it really well can get you very far.