A Hundred Users Signed Up and Never Came Back

Lex Fridman opens with the fact that Armstrong is a programmer, and asks for the first program he remembers. In middle school, during a recess reading period, the other kids had comic books. Armstrong had checked out a library book he remembers as *How to Learn Java in 30 Days*. He understood none of it. At home he tried to get a program running and hit Java's first lines — `public static void main(String[] args)` — which felt foreign and hostile.

The first thing he wrote was probably a Hello World in Java, but he was so confused about what was actually happening that he later learned PHP. PHP was more fun: you tell it to print something and it prints it, without the ceremony. He built simple websites, picked up HTML, and still feels he has to keep a fondness for PHP as a dirty secret. He is more open about Ruby, still his favorite language. The idea behind Ruby, he says, is to make it easier for the human and harder for the computer, and to make being expressive a joy.

He was never the best computer scientist. He was a good hacker who could rapidly prototype products, and languages like Ruby were the tool for that. Fridman notes that Java's Hello World is among the longest you can write, which is awkward given how long universities used Java to teach programming. Armstrong later wrote an entire Bitcoin node in Ruby — a weird decision in hindsight, because Ruby was not the most performant language, and Coinbase has had to rebuild that stack many times.

Fridman asks the practical question: what is a cryptocurrency exchange, and what does Coinbase actually do? Armstrong's answer is that it is an exchange, a brokerage, and a custodian: the primary financial account for people in the crypto economy, covering how they buy crypto, store it, and use it. The hope is that a billion people will access the open financial system through it. Under the hood an exchange keeps an order book of bids and asks; a trade executes when someone is willing to buy for more than the lowest offer. A brokerage is simpler still: you say you want a hundred dollars of Bitcoin, you get a quote, and you hit accept.

That one-second experience sits on a pile of unglamorous work. Coinbase has to plug into payment rails around the world so people can get fiat into crypto. It has to fight hackers trying to steal crypto or feed stolen cards and bank accounts into the system. It has to keep up with blockchains that get updated, fork, and airdrop. And it has to store the private keys of what customers buy. Armstrong says the custody technology is what helps him sleep at night as one of the largest crypto custodians. The company now has a portfolio of products — institutional trading, Coinbase Commerce as "Stripe for crypto," a self-custodial wallet, and an NFT product — but most of the revenue is still people coming in to buy and sell.

Fraud prevention is a cat-and-mouse game with no silver bullet, only lead ones: test suites, third-party audits, a bug bounty for white hats, and machine learning over labeled good and bad actors. Relatively well-known signals include device fingerprints (fonts, emulators, virtual machines), a metric Armstrong liked called improbable travel velocity — an IP in Austin and another in London an hour later — and the milliseconds between keystrokes, because real users type a card number one digit at a time while scammers paste a whole list. High-frequency traders have also fired a million API quotes looking for one that is out of bounds. When Coinbase gets the math wrong, someone arbitrages it, the company loses money, and then it fixes the hole.

Armstrong started Coinbase when Bitcoin was about ten dollars. He was a software engineer at Airbnb, working on fraud prevention, when he read the Bitcoin white paper in December 2010 and started going to Bay Area meetups. SMTP needed Gmail; git needed GitHub; most people do not want to run their own email or git server. His hunch was that Bitcoin would need a hosted wallet that handled security and backups. Smart friends who did not understand Bitcoin thought it sounded like a scam. Friends who did understand it thought storing everyone's coins was a way to get hacked. So he built a nights-and-weekends prototype, labeled it a beta, and told people not to put real money in it.

He was 29, almost 30, and wanted to be an entrepreneur. He applied to Y Combinator hoping a check would make it feel like a real company, and he could not find a co-founder. He describes that stretch as wandering in the desert with a lot of self-doubt: maybe Bitcoin would get shut down, maybe the whole thing was stupid. Paul Graham and the YC group wrote a check for about $150,000 after the interview. That was the first time someone he looked up to said the idea was worth pursuing. It gave him the confidence to quit Airbnb.

He found the right co-founder after YC. They still had no customers. He posted the hosted wallet on Reddit, maybe a hundred people signed up, and nobody came back. YC's advice is to talk to customers, so he emailed five of them and got on the phone. Why didn't you come back? One man said the app was okay for a beta, but he did not have any bitcoins, so he did not know what to do with it. Armstrong asked whether a buy-Bitcoin button would have helped. "Yeah, maybe." They then had to assemble bank partnerships, payment rails, and a basic exchange. The minute they launched the feature — click buy, put in a bank account or credit card, Bitcoin shows up — users started going up and did not stop. After two years of wandering, they had product-market fit.

Fridman asks about the darkest moments. Armstrong had a chip on his shoulder: he wanted to do something important, and he was willing to sacrifice sleep and nights out for it. Coinbase had maybe five employees. The site went offline in the middle of the night, and he was on PagerDuty, woken two or three times a night. Sleep deprivation, he notes, shows up in disasters like the Exxon Valdez spill because it magnifies everything else. Then a bug made balances display wrong. Thousands of people on Reddit and Twitter piled on: the company is over, give the money back. He had never had a thousand people mad at him at the same time.

Customer support was backed up around twenty thousand requests. Someone posted his cell phone number on Reddit so angry users could reach the CEO. It was late, they had been working twelve hours, they needed food delivered, and the phone had been blowing up all day. He started answering calls to catch the delivery driver. "Is this Brian?" Wrong number, click. Next call: a reporter from Japan asking about a security issue. Wrong number, click. He finally got the food upstairs. At some point that night he curled into a ball on the floor and cried. He let himself wallow in self-pity, maybe took a five-minute nap, then told himself to solve the problem and got back up.

The source of strength was a co-founder. Armstrong was lucky to find Fred Ehrsam; there were moments at wit's end when Ehrsam rallied the team and carried it. On the broader problem of public hate, Armstrong's view is that almost anything significant will anger some slice of people, maybe five percent, maybe forty-nine. You cannot live in a bubble of yes-men, and you also cannot try to please everyone. He has noticed that some leaders read as PR-rehearsed while Elon Musk says whatever he thinks, to a fault — and even people who hate what Musk says at least believe it is authentic. The internet puts a premium on that authenticity, which is a hard line for a public-company CEO to walk.

Advice for the five-employee stage, pre-product-market fit: action produces information. Armstrong cites Paul Graham's line that startups are like sharks — if they stop swimming they die. Even if you are not sure what to do, do something, because shipping produces a signal. There were times he built the wrong thing, and the minute it shipped he knew what to do next, an idea he would not have had if they had kept debating. His other analogy is a mountain shrouded in fog: you can only see three or four steps, you take them, and the next three appear. Most people never step into the fog because they fear failure, running out of money, or not getting a job afterward.

He thinks that fear is overstated. If a startup fails you are more valuable to the next employer. You can raise a seed round, pay yourself a salary, try for two or three years, and go get another job if it does not work. You were paying yourself during that time. People overestimate the risk and never start, because the default reaction to a big idea is that your friends think it is silly. You know you have product-market fit when usage keeps growing without marketing dollars. Until then you watch stats, see false-hope wiggles, talk to customers, improve the product, and try not to run out of money.

Post-fit, the problems flip to scaling, hiring, and not dying in a downturn. An investor told Armstrong to invert his time: he was still writing code maybe fifty percent of the time and hiring twenty, and that would make him the bottleneck. He still has to push decision-making down in a five-thousand-person company. Early finances were a mess; after a Series C he parked about $25 million in a separate bank account the operating business could not touch, a cheap hack until they hired a real CFO. The first hire was customer support, then engineers, then around hire six a recruiter as a force multiplier. Legal and compliance came because nobody had more than a year or two of crypto experience.

Coinbase's mission is to increase economic freedom in the world. Armstrong treats the phrase the way economists treat GDP: a scored bundle of property rights, free trade, a stable currency, the ability to start and join companies, and the absence of endemic bribery. Those scores, he says, correlate with higher growth, higher self-reported happiness, better environmental treatment, and better incomes for the poorest ten percent, and they negatively correlate with corruption, bribery, and even war. Give people property rights, rule of law, and trade, and they do more good things.

He spent a year in Buenos Aires after college, originally planning a month in each South American city, and stayed once he had an apartment and a phone. Argentina had been through hyperinflation. He describes a pessimism that creeps into countries without economic freedom: don't stick your head up, don't try too hard, because it could all be gone tomorrow, and the valuable things in life are family, friends, and a past that was better than the future. In high-freedom countries, people try more stuff because they expect to keep some of the upside.

Reading the Bitcoin white paper, he had a hunch that this technology could inject good financial infrastructure — property rights you can hold as a twelve-word phrase on a phone — into countries that lacked it. Refugees can take wealth with them; cross-border payments do not require a bank in every jurisdiction. Bitcoin's twenty-one million cap is, in his framing, a digital return to the gold standard in an era of fiat that eventually gets printed. Decentralized finance goes further: Uniswap is a smart contract on Ethereum with no company to shut down. Someone in India can borrow from someone in the United States without collecting licenses in two hundred countries. Fridman asks whether Uniswap is a competitor. Armstrong does not think of it that way; Coinbase has integrated Uniswap and wants more activity to move on-chain over time.

Coinbase's brokerage is a centralized, regulated financial service, and Armstrong is proud of it because that is how fiat flows into crypto. Once people have crypto, they can hold it in a self-custodial wallet, which is more authentic to the ethos. Custodial products meet people where they are: simpler, less fear of a user error wiping out savings. Self-custodial means the keys live on your device. Even if Coinbase gets a court order to seize funds, from an architecture point of view it cannot. If Coinbase gets hacked, it cannot lose those funds. The cost is that responsibility moves to the individual.

A lot of cryptography research is trying to make that responsibility survivable. Multi-sig can require two of three keys, or three of five. Multi-party computation and threshold signatures are in the same family. A simple picture: one key on your phone, one at Coinbase, which is not a quorum by itself, and a third in iCloud, Google Drive, or another party that is not Coinbase. Day to day, phone plus Coinbase signs instantly. Lose the phone, and Coinbase plus the backup recover a new key onto a new device. No single party can move the money.

Fridman presses the business question: doesn't self-custody bleed transactions and make it easier to go somewhere else? Armstrong says that is the point. A traditional financial firm would hoard custody to look valuable. He wants users who are ready to move to self-custody. Everyone is using the same protocol, so switching costs are low, which keeps companies accountable. If you want the Visa network, there is one company in the world. If you want Bitcoin, there are dozens or hundreds. It is arguably worse for Coinbase as a company and better for customers. On-chain versus off-chain is a related tension: early on, moving between Coinbase accounts off-chain was a hack to avoid miner fees; as L2s scale, more should go on-chain, though high-frequency traders still cannot put everything on a public chain.

Coinbase lists more than a hundred assets, depending on jurisdiction and whether the customer is retail or institutional. Armstrong wants to get away from "listed on Coinbase" as an endorsement. Imagine, in the early internet, a committee deciding which web pages were good; it would have been wrong, the way Apple's App Store review process keeps getting in trouble. The listing bar is a legality test: if they believe an asset is a security under current U.S. law, it cannot be listed, though they acquired a broker-dealer license and hope someday to trade crypto securities. They also look at smart-contract security — can someone move customer funds without permission? — and at the people behind the project, including criminal history. If it clears that minimum bar, they want the market to decide, Amazon-style: three stars or five is fine; consistent one-star fraud might get removed.

He expects millions of assets over time. Low-market-cap coins will trade on DEXes through Coinbase, which makes it even clearer that listing is not a blessing. He is cautious about the word "scam" because many cases are judgment calls a court or jury could go either way. Red flags include a large insider allocation with short vesting, founders with criminal records or prior frauds, a me-too product with no real innovation, and marketing that pumps the future price on YouTube and Twitter. The useful fact is that much of this is now on-chain: you can see who owns the tokens and whether they are selling.

Privacy coins are the hard case. Fully anonymous assets like Monero make it difficult for a licensed firm to run a defensible anti-money-laundering program, because you cannot do blockchain analytics on the source and destination of funds. Zcash is a compromise: it has a view key that can de-anonymize transactions in specific situations. Coinbase supports Zcash because purchases on the platform are not anonymous and the firm can see where the coins go afterward. A few hops later a user might turn privacy back on — a tough judgment call. Then Armstrong takes his Coinbase hat off. He thinks privacy coins are good for the world and should be allowed. The early internet was all HTTP; people were afraid to put credit cards on it. Now the web is HTTPS with a lock icon. Financial information is some of the most important information to keep private. A public ledger plus an identity match de-anonymizes your whole life. Charities may want full transparency; personal money should not have to broadcast itself.

Coinbase has chosen a seat at the table with regulators. Armstrong's trend line is that five years ago the common question was how to prevent all this bad activity; today it is more often how to preserve the innovation while keeping bad actors out. In D.C. specifically, he puts fifty to sixty percent of the people he meets in the camp of "this has upside, regulate the bad uses, keep the work in the United States" — almost double the thirty percent of three years earlier. True crypto-skeptics willing to call it a net negative are maybe twenty to thirty percent, and that position is getting harder to defend because roughly one in five Americans have used or tried crypto. Calling it all bad is condemning a fifth of your fellow citizens. Dubai, the U.K., and Australia are competing for startups with clearer rules. China treats crypto as a threat to power.

What he wants is a classification test, because crypto is not one thing. Bitcoin, Ethereum, and many others look more like commodities, not controlled by any one person, and belong at the CFTC. Raising money for a company looks like a security, for the SEC. Stablecoins and central-bank digital currencies look like currencies, for Treasury. NFTs, metaverse items, decentralized identity, and voting may be none of the above. A simple threshold he offers: if nobody controls more than twenty percent, it is probably more like a commodity; if someone is selling it to fund a business, it is probably a security. He also wants a sandbox so a startup below some payment-volume or custody size is not crushed by legal bills before it can get off the ground. Pension funds and sovereign wealth funds tell him they hold about one percent in crypto and would rather hold twenty, waiting on that clarity.

Can a government seize crypto by calling Coinbase? That, he says, is why people want self-custody. In the United States there is rule of law and protections around search and seizure. Coinbase publishes transparency reports, receives subpoenas and court orders, and has been ordered to freeze accounts; as a regulated business it has to follow the law. It has also gone to court over overly broad requests — freeze this huge number of accounts — and won some of those cases on behalf of customers. At a certain size, armies of lawyers become a tax that gets passed into fees. For a Coinbase lawyer the gray is both hard and interesting: there is little case law, which is how legal careers get made.

Asked whether Bitcoin or some other crypto will become a world reserve currency, Armstrong says yes, controversially. He had been reading Ray Dalio's *The Changing World Order*, which tracks how reserve currencies rise and fall with empires. The United States, by some measures, has had a good run and is coming down a little; China is coming up. If the dollar sees more inflation, the yuan is not obviously better — China has a lot of debt too, and the yuan can be printed. What he watches instead is a cohort of roughly 25-to-35-year-old, tech-savvy people who already treat crypto as the primary store of wealth and convert to dollars or euros only at the last minute, the way a traveler converts local cash. That group is not huge in GDP terms yet. He reads it as a leading indicator, and as a geopolitical split: a more centralized East versus a West that embraces an open, fair, global financial system, with Bitcoin as the gold standard of the crypto economy.

Fridman then turns to Armstrong's September 2020 blog post, "Coinbase is a mission-focused company." Armstrong meant that great companies have an important mission and were getting distracted by employee activism that pulled the company into whatever the current thing was. He thinks of it as largely an American phenomenon and worries it makes the country less competitive. Employees outside the U.S. were confused: you're just saying we focus on work at work? Inside certain U.S. cities, people wanted the company to act almost like a government, with an opinion on nearly every hard societal issue. Questions to the CEO shifted from products, competitors, and regulators to "what's your stance on X." Internal Slack turned into social media. It culminated in a walkout and a list of demands. Armstrong decided that if the CEO job was to squirm daily in front of the hardest social questions, he did not want it. He had founded the company. People who were not on board could leave.

He offered an exit package; five percent of employees took it. Political and social activism outside of work was fine. Disagreement about the work itself was also fine — that is how you get to the truth. What he wanted out of the office was division that had nothing to do with the mission. He knew he would be attacked. Mainstream outlets wrote hit pieces; he says he lost a lot of trust in that media, which he now sees as hyper-politicized and not that focused on truth. Internally there was a large outpouring of thanks. He cites a Nassim Taleb idea about the tyranny of a small intense minority: maybe five percent are really upset, another fifteen percent go along because the cause sounds reasonable, and eighty percent want to work without the drama but are afraid to speak up. People warned that underrepresented groups would never join Coinbase again. He asked employee resource groups and heard they wanted respect and good work. A year later, diversity numbers were the same or better in every category. The hiring upside was engineers from places like early Google who wanted a company willing to stand for focus.

Armstrong never thought of himself as a natural leader. The picture in his head was a military general, confident, barking orders. He is more introverted, not that confident in how he communicates, and he prefers a product-and-technical CEO style: hear everyone, then send a decision later that day rather than rendering one in a heated room. He watches whether people are trying to be right or trying to seek the truth, and he uses the trick of making each side argue the other's position. Never make a decision when you are angry, because judgment drops. Confidence, for him, is just doing things a little outside the comfort zone until the zone gets bigger. He had never managed anyone when Coinbase started. Publishing the mission blog was scary: he did not know if one percent or fifty percent would leave. Every year or two the company still produces a "next level" that costs him a week of sleep.

After Coinbase went public he had some liquidity and funded ResearchHub, an attempt to make scientific research more like open source software. The funding process, peer review, journal submission, and journal costs all feel antiquated: scientists often are not paid for the paper or the review, the work sits behind a paywall, and taxpayers pay again to read it. Preprint servers like arXiv and bioRxiv helped, but the sites look fifteen years old. ResearchHub's first version is Reddit for science: hubs as journals, papers you can publish, an electronic lab notebook so a paper is a living document rather than a static PDF, plus code and datasets. The longer bet is funding and licensing, because scientists and company-builders rarely overlap, and when they do you get SpaceX, Genentech, Google — even Coinbase, which grew out of a white paper. Most businesses have no scientific innovation; a lot of science never becomes a product. He wants a translation layer, and a new prestige economy: monthly and yearly leaderboards, grants, a ResearchHub prize, even a print issue you could hang in a lab. "Friends don't let friends publish in paywall journals." A token, Research Coin, accumulates when people upvote a paper, a form of reputation-weighted peer review. YouTube started as cat videos; a million subscribers now beats a show on NBC. He hopes in ten or twenty years young scientists will see open publishing as the prestigious path.

NewLimit is a longevity bet. Hosting dinners with scientists, he got interested in cell reprogramming. Shinya Yamanaka's 2006 Nobel work showed you can turn a skin cell back into a stem cell; going all the way back risks cancer. The idea is transcription factors that make a cell act a bit younger without fully de-differentiating. The moonshot is a therapy, maybe in ten or twenty years, that rejuvenates tissue across the body, even the brain, so older people are less ossified in their thinking. He says the impulse is a bit Elon Musk: pick a high-tech-risk problem that would matter enormously if it worked. Founders who made money in software's golden age should do atoms, not just bits — biotech, cars, rockets. NewLimit was a handful of people and was hiring scientists interested in reprogramming, not necessarily from aging. ResearchHub needed software, design, and product help.

Advice for young people: there is no one way to be happy, but a lot of people want impact. Armstrong thinks too many of them conclude that the path is to be an activist, to speak truth to power and pressure the people who have it. Criticism is easy and wins accolades from friends. Changing the thing is hard. He points to Theodore Roosevelt's arena: go chew glass and stare into the abyss. If the problem can be a company, join one or start one. If not, start a charity. Be part of the solution. Technology, in his view, is still one of the most important ways to improve the world — carbon sequestration, Khan Academy, transportation, the financial system — and the anti-tech mood confuses him. New tools get abused by some people; he still thinks most people are good and most of these tools are net positive. If you are not sure what to do, start anything. That is how you learn.

On the meaning of life he is blunt: he does not really think there is a point. A friend told him that if you go too deep into the big existential questions you stare off a cliff and there is nothing there — snorkel, don't scuba. As far as he can tell, organic molecules in the ocean started dividing, the selfish gene did its work, humans grew neocortexes, and now we ask these questions. He just wants to keep watching the movie unfold. That is part of why NewLimit appeals: if people live longer, or upload a brain, or strong AI works, we get more of the film. Fridman adds that we also get to write the movie as we watch it. Armstrong agrees, citing the Steve Jobs point that everything around you was invented by somebody who had a crazy idea.

In his twenties he read self-improvement books and wrote down goals: make a hundred thousand dollars a year, own rental property. Those happened over a couple of years, so he thought bigger. The craziest thing he could write was "start a billion-dollar tech company," which was absurd for someone who had never started a million-dollar one. He wrote it on paper almost every day for a year. Small things started to happen: maybe move back to the Bay Area from Buenos Aires, maybe apply to Y Combinator. Within about ten years Coinbase's valuation was over a billion dollars. The lesson he kept is that you can accomplish more in ten years than you think, and less in one year. The next goals on the page are a billion people accessing the open financial system through Coinbase every day — one in eight humans — or a serious extension of human healthspan. He does not subscribe to the woo version. The rational version is that if you write the long goal every morning, it is on your mind, so you notice opportunities you would have missed.