Nobody Serious Believes Cookie Banners Help Anyone

Lex Fridman opens by noting that DHH, for someone now treated as a legendary programmer, got started late — and only after two failed attempts. The first computer he sat in front of was a neighbor's Commodore 64 at age five, playing *Yie Ar Kung-Fu* with a pile of kids in a Copenhagen bedroom. He begged his father for one; the machine that arrived was an Amstrad 464. Magazines in 1984 printed game source in the back, and he would spend two hours typing a listing, misspell a word, and watch the whole thing fail. Constants made sense to him. Variables did not. Then he discovered piracy, and the urge to learn went quiet.

The second try came around age 11 or 12 on an Amiga 500, still his favorite computer, with a language literally called EasyAMOS. He got conditionals and loops and still could not finish even a basic game, and he decided he was not smart enough — not the math kid. What he *could* join was the European demo scene: groups competing to squeeze near-music-video demos into four kilobytes, swapping 3.5-inch disks by mail. At 14 he ran a BBS from a tiny bedroom with three phone lines and an Amiga 4000, trading demos and pirated software in a Danish culture that, he says, never even staged the argument that piracy was wrong. He once spent a month of newspaper-delivery wages on *After Burner II* for a Sega Master System — 600 kroner — and never waited that long again.

The click came on the third attempt. In ninth grade a university excursion sat the class in front of Netscape Navigator and a text editor; he made text blink with an HTML tag and felt, for the first time, a success that did not require an error message. Gaming-review sites led to ASP, then PHP, and PHP is where loops and variables finally composed into applications he could ship. He calls late-1990s PHP the pinnacle of web-developer ergonomics: write a script, FTP it, reload. No framework, no build, no ceremony. He has been chasing that high ever since, which is why he lights up at Pieter Levels still running businesses on PHP, jQuery, and SQLite. The industry, he argues, handed too much of that feeling to "the merchants of complexity." Rails 8's no-build push is his attempt to get the 1990s ease back without pretending the 1990s were better in every way.

JavaScript is DHH's distant second-favorite language after Ruby, and he does not hate the language itself. What he hated was the 2010s toolchain: webpack projects that would not compile if you put them down for five minutes, frameworks rewritten every quarter, Hacker News threads treating last season's stack as a moral failure. He calls that stretch the dark ages. The browsers of the mid-2000s had been frozen after Internet Explorer 5 won the war and Microsoft, in his telling, disbanded the team. Firefox kindled something; Chrome and Google's dependence on an open web actually moved the platform forward. A lot of the complexity that followed was a bridge, he grants — not a destination. The mistake was treating the bridge as housing.

Some of that complexity, he says, came from org charts. At a company like Facebook the web developer was no longer one person but a stack of niches, and complexity followed the slices. Some of it was necessary to get browsers to where they are now, when you can write a text file, serve it, and get a serious application without a preprocessor. He is almost in awe that a ninth-grade webpage from 30 years ago still renders, and that building a new engine is an Apollo-scale project — which is why he funds Ladybird and still credits Chrome as "a complete gift to web developers." The spying and ads he will criticize; the browser he will not.

Asked about the U.S. Department of Justice's push to split Chrome from Google, DHH does not hedge: it is a disaster. He is sympathetic to antitrust in tech, but browsers are the wrong target. Firefox still exists, Safari still exists, and nobody is forced through Chrome to use the open web. The real toll booths, he says, sit on mobile phones. If you want to go after Google, go after the ad ledger where it sits on both supply and demand — not the engine that kept the web from becoming a set of native stores Apple and Microsoft would have been happy to own. Chrome, he adds, did not win by locking Android; unlike iOS, Android still lets you ship a real alternative engine. Edge is Chrome in a dress. DuckDuckGo, the search engine he uses, used to dress itself in Bing.

The cautionary tale he wants legislators to remember is GDPR. A decade or twelve years on, the measurable result is cookie banners that everyone hates, that save no privacy, and that have "only enriched lawyers and accountants and bureaucrats." He calls them a monument to good intentions leading straight to hell, and says Europe is world-class at that particular craft. The internet got uglier in one stroke. Worse, everyone has known the failure for a decade and still cannot repeal it. That, to him, is a bankruptcy declaration for any bureaucracy that pretends to make life better.

The punchline is the one Lex used as the episode highlight. It is not just the EU. You cannot hide from cookie banners anywhere on the planet. "If you go to goddamn Mars on one of Elon's rockets and you try to access a webpage, you still see a cookie banner." Lex jokes about the 150-second ping; DHH estimates 45 seconds just to click through. Nobody in the universe is safe from the nonsense. The rest of the conversation is the same test applied to languages, clouds, app stores, and families: never mind the intention — did the thing actually make anyone's life better?

PHP taught him to ship, but he never fell in love with it. The love affair started in 2001, when Jason Fried hired him over email — Copenhagen to Chicago, a kid answering a PHP question on the internet — and they still work together 24 years later. Client work kept dying in email threads and `finalfinal_v06` FTP folders, so they built Basecamp: one place for the project, simple enough that nobody needed a seminar. For the first time DHH had free choice of tools. He had been reading Dave Thomas and Martin Fowler in IEEE using Ruby as pseudocode. Matz had started the language in 1993. DHH gave himself two weeks to talk to a database and render HTML. It took a weekend. "Ruby was made for my brain like a perfect tailored glove by someone I'd never met."

The aesthetics are the point. Integers are objects, so `5.times` is a loop with almost no line noise. Python's `__init__(self)` offends him as the most important method in the ugliest possible spelling; Ruby's `initialize` refuses abbreviations that only help the parser. `if user.admin?` uses a question mark that means nothing to the machine and everything to a human. `user.downgrade unless user.admin?` deletes the exclamation-point clutter of a negated if. All of those forms compile to the same thing. They exist because Matz's number-one goal was programmer happiness, which DHH contrasts with James Gosling's darker view that average programmers cannot be trusted with sharp tools. Matz even lets you extend base classes: DHH added `5.days` via Active Support so cache expiry reads like English. A stranger from Denmark was allowed to write new chapters that readers cannot tell from Matz's own.

Metaprogramming turns that trust up to eleven. Active Record lets you write `has_many :comments` and `belongs_to :account` as if they were keywords; they are domain-specific language sitting on Ruby, with none of the parentheses pile-up he associates with Lisp. He has never written real Lisp, he admits, and judges it by looking at the line noise. Smalltalk he respects as an experiment. The first drafts of Ruby, he notes, were terribly slow; hundreds of person-years later the language is both beautiful and fast, and about 90 percent of what people use in 2025 was already in that first draft. He also has a mischievous open-source habit: ship the first version a little broken, so collaborators have something obvious to fix.

DHH will defend dynamic typing to the death because it *is* Ruby: duck typing, runtime methods, metaprogramming that does not look like TypeScript gymnastics. He pulled TypeScript out of Turbo after trying to write metaprogramming in it. The stereotypical `User user = new User()` repeats the same word three times, and he does not have time for that. He does not write code with tools, he says. "I chisel them out of the screen with my bare hands." No autocomplete, no IDE, on purpose — because the moment boilerplate is free to type, you tolerate a fabric full of it. He catches bugs with tests, not with a compiler. He fully accepts that a 10-million-line codebase with tens of thousands of programmers needs different methods. What he refuses is the idea that the same methods belong in a bedroom building Basecamp. "You don't use your Pagani to go pick up groceries at Costco."

Shopify is his exhibit A. On Black Friday it has done on the order of a million dynamic requests per second, running something like 30 percent of internet e-commerce stores. The monolith is about five million lines of Ruby; the same system in Go or Java, he guesses, would be 25 million. The company has just under 10,000 employees, a market cap around $120 billion, and quarterly GMV on the order of a quarter of a trillion dollars. Tobi Lütke was on the Rails core team, wrote Active Merchant and the Liquid templating language, and still sweats details as CEO. DHH does not design Rails for Shopify-now. He designs it for Tobi writing Snow Devil alone — the one-developer snowboard shop that became the company — because the runway from FTP-era beginner to world commerce is the argument.

Twitter leaving Rails, he says with a grin, became the best advertisement the framework ever had: after the rewrite, "nothing fucking happened for 10 years." Scale, unpacked, is two packages people mash together. Latency is one: he wants requests under 300 milliseconds, ideally around 100. Horizontal scale is another, and languages all add boxes. The hard part is MySQL, not Ruby. YJIT is Shopify's gift to the ecosystem, wringing out five or ten percent at their scale, but they did not *need* it to get huge. The real question is economic. 37signals once spent about 15 percent of operating budget on Ruby app servers; doubling language speed would move the business by a few points, while 10 percent more programmer productivity would move it far more. Ruby is a luxury language — "the Coco Chanel of programming languages." Some requests are too cheap to afford it. Most business software is not. The expensive cores are wet cores: humans who have to understand the system.

Lex likes collaborating with AI and argues that a human-readable language like Ruby is a high-bandwidth interface for that pairing. DHH agrees on the bandwidth: Elon's Neuralink riff on the podcast was that speech is low bit-rate, and Ruby packs more concept per character. He uses AI constantly, in a separate window. He has tried Cursor and similar drivers and does not enjoy them. "I can literally feel competence draining out of my fingers." Building Omakub, his Ubuntu remix, in Bash, he asked the model for the same conditional so many times that he used the language without learning it. You learn with your fingers, he says; watching guitar YouTube does not teach guitar. Claude Code in the terminal also feels like waiting for a compile, and after 20 years of Ruby he no longer has compile-wait in him.

The deeper fear is the same thing that happens when great programmers get promoted into meetings: they lose touch. He does not want to become "a project manager of a murder of AI crows." He could have become a project manager 20 years ago if outcomes were the only value. Programming for him is also the playing of the instrument. Vibe coding is fine when he just wants to see an idea; it does not make you a programmer, any more than Excel made accountants into software engineers — a wonderful expansion of who can explore ideas, not a substitute for the craft. Editing, he insists, is the reward for being a doer. Prompt engineering as a wizard class is a pipe dream; the beautiful promise of AI is that a clumsy sentence from someone who does not know the machinery still produces something they want.

Used as a pair-programming buddy that never judges a dumb question, AI makes him about half a percent smarter at the end of a day. Used as a driver on Omakub, he felt half a percent dumber. He is more than a little skeptical of the Anthropic CEO's claim that 90 percent of code will be written by AI by year's end, and he recites his first rule: "nobody fucking knows anything." DeepSeek's local think-box scared him more than ChatGPT. *The Lawnmower Man* promised VR in five years; 30 years later his kid plays *Gorilla Tag*. Airplanes went from the Wright brothers to jets in four decades and then flattened. COBOL still cuts Social Security checks. His advice to a young programmer is therefore almost stubbornly simple: if you like programming, learn it by programming. It may become a horse — recreation, not transport — and that would be okay. He has already had decades of being paid millions to write Ruby. Gratitude, not regret. Position yourself so that if you have to retool, you will not regret the path.

In 2012 DHH wrote the Rails doctrine because communities that do not record their values eventually fill with people who have other ideas, and then someone yanks Chesterton's fence and the wolves come in. Programmer happiness is first, homage to Matz, including ambiguity: IRB accepts both `exit` and `quit`, while Python lectures you to read the manual. Convention over configuration was born from Java XML hell; pick `post_id` and stop asking people to assemble the Millennium Falcon from a box of Legos — hand them a finished toy they can still take apart. The menu is omakase: Rails tries to solve the whole web problem with a chef's menu, against a JavaScript ecosystem where nobody owns the pairing of controller, database, and the rest, so wisdom does not accumulate. No one paradigm: object-oriented at the core, with functional and imperative borrowed in. Beautiful code is a goal, not an accident; a `belongs_to` / `has_many` / `validates` model reads, Lex says, like a haiku.

Provide sharp knives, not Java's butter knife. Value integrated systems. Progress over stability is the one he would rewrite: Rails 8 was essentially a no-op upgrade from 7, and the JavaScript 2010s taught him that churn as a value is an indignity. The hidden tenth commandment is the monolith. Microservices can make sense at Netflix scale; applied to 20 programmers and half a million lines, "you're an idiot." First rule of distributed programming: do not distribute your programming. Basecamp and HEY are each about 100,000 lines of Ruby. Basecamp has something like 420 screens. HEY launched with 40 kilobytes of JavaScript against Gmail's roughly 28 megabytes uncompressed, six compressed. Splitting front end from back end, he says, was one of the great crimes against web development.

Active Record is the crown jewel, named without shame after Martin Fowler's pattern in *Patterns of Enterprise Application Architecture* (2001), a book he has read many times. Tables become classes, rows become objects. It does not hide SQL; it makes the tedious parts less cumbersome so you can still learn the query language — and you should, even with an ORM. SQL survived the NoSQL decade and still reigns. Other ORMs tried to live in a pure object world and then invented a translation layer; Active Record admits the record is going into a SQL database and structures itself around columns. The soft ramp of Rails is the same as Ruby's: generators can put a blog on screen before you understand anything, as an invitation to competence, not a claim that competence no longer matters. That is also his quarrel with the AI meme that programmers no longer have to know things.

He once doubted his own allergy to engineering managers, hired good ones, ran the one-on-ones for a couple of years, and concluded he had been right. Not every programmer needs weekly therapy with a manager. The best feedback on the work comes from someone better at your job than you are, and former programmers who stop coding turn into pointy-haired bosses remarkably fast. John Carmack's id Software, in DHH's memory, was a handful of designers and programmers making *Quake II*. Gmail began as a Paul Buchheit side project. 37signals went 19 years without a full-time finance person, treating the books like an ice cream shop while taking in hundreds of millions in revenue; a two-year, $5 million sales-tax mess finally bought them a head of finance, who replaced a decade of full-time analytics. If he has to choose, he picks the accountant.

The default team is two: one programmer, one designer, one feature. You cannot make Basecamp with a thousand people; you will make thousand-person software. Humans do not scale, and communication scales worse. Small is not a stepping stone — that was a thesis of *Rework* — and the protection is simple: do not take venture capital, because other people's money wants a return, usually via the enterprise-sales playbook that turns a product into a thousand-person machine. Jeff Bezos bought secondaries from Jason and DHH personally, a few million each, not a growth check into the company. His investment team said the price made no economic sense. Bezos said he did not care; he wanted to invest in this guy. The money was a vaccine against a larger VC check. Once a year he would get on a call and tell them to keep doing awesome stuff, because he did not know how to run their business. DHH looks back with some regret at taking that vote of confidence for granted.

Jason Fried and DHH have been partners for a quarter century, living in different cities for most of it, talking less than two hours in a typical week. Early product fights were so loud that employees thought the company was dying; inside the room it was never personal — "Jason, you're an idiot" because the idea was wrong. Distinct competence is the glue: Jason owns design and product, DHH owns tech, and on the overlaps they use Bezos's disagree-and-commit. They like doing the work, not living in the idea layer. "Meetings are toxic." Founders who promote themselves into full-time idea people never have to implement, so they never slow the horse. For loneliness on remote work, DHH's unsolicited advice is to get a wife and kids. Family, 40-hour weeks, hobbies, and friends are how you last 25 years instead of sprinting toward an exit you will hate. Mojito Island is a mirage. Basecamp not being Jira is not a failure; hundreds of millions of dollars kept, with a life attached, is a path Silicon Valley still treats as forbidden knowledge.

Amazon, like any company that actually changed commerce, must have a black-hole-sized population of haters or the universe would tip over. DHH credits Kathy Sierra's *Creating Passionate Users* with carving that physics into his brain: value that many people love creates an equal and opposite force of people who hate you. The only escape is mediocrity. He applies the same ledger to Elon later in the conversation. The point is not to collect enemies. It is to stop treating controversy as evidence you did something wrong, and boredom as evidence you did something safe.

The case against retirement is the same physics from the other side. Ambitious people do not sip mojitos for long. Mihaly Csikszentmihalyi's *Flow* — a name DHH practiced with AI so he would not butcher it on air — found that people are happiest stretching just past current skill, losing three hours that felt like 20 minutes. Take away the problems and those people get depressed. Selling the company and sitting still is hell; failing after a sacrificed decade is hell; staying and working only is also hell, because it never gets easier. He wanted flow for 40 or 50 years, which meant not betting the entire stool on work. Lightning in a bottle rarely strikes twice, and he would tell his younger self nothing, because ignorance of how hard it will be is half the blessing. Responsibility, via Jordan Peterson and Viktor Frankl, is the reason hardship is bearable. You do not get your 20s, 30s, or 40s back.

Derek Sivers's bicycle ride is the bargain he wants: all-out in 19 minutes, or 5 percent slower in 21, two minutes apart, with birdsong. Being the best in the world at one thing requires mania — Senna, Michael Jordan — and he knew early he was not that person. He failed EasyAMOS. The cheaper bargain is the top 5 percent, which he figures is 20 or 100 times easier than the top 0.1 percent, and you can hold it in five things at once: writing, racing, programming, a company, a family. Karl Marx's morning fishing and evening hammering is, for him, a diagnosis of alienation he cannot stomach. After weeks deep in code he has to come up for air, annoy people on the internet, or drive, and then the programming comes back at full intensity.

The internet meme is "DHH left the cloud." The critic's version is that 37signals wasted time pinching pennies instead of going to the moon with VC and AI. Opportunity cost is real, he says. Saving about $2 million a year — half to two thirds of the infrastructure bill going to the bottom line and profit sharing — is still worth doing at their size. Silicon Valley's "costs don't matter" ethos is, to him, aesthetically ugly, a splinter like line noise in a program. HEY launched entirely on AWS. The peak bill for the company was about $3.2 to $3.4 million a year. The pitch had been easier, cheaper, faster, with Jeff's line about a competitor's margin as your opportunity ringing in his ears. AWS, he notes, runs around 40 percent margins. The pitch, in his optics, was false.

It did not get easier: IAM rules are worse than Linux. It did not get cheaper: they cut spend by half to two thirds without hiring anyone. Faster is true if you need a thousand machines in 15 minutes; if you need 20 servers you can wait a week, buy more metal, and skip the auto-scaling religion. In early 2023 they moved seven major applications off cloud compute, cache, and databases in a little over six months, onto a fleet from Michael Dell, 41 years in the server business. S3 stayed under a four-year contract, because decent cloud pricing means binding yourself for years anyway. He projects on the order of $10 million saved over five years. US-East-1 taking a third of the internet with it is, he says, an insult to DARPA's design of a network with no single node to kill.

Owning hardware has pain — computers are painful — but the expertise is not nuclear rockets. The whole internet was ethernet, power, Linux, until the cloud took over a decade ago. They do not drive to the data center; Summit, formerly Deft, is white-glove, and what they see is IPs and drives coming online, like the cloud at half the price. Homelab is the next itch: five-gigabit home fiber, NVMe making disk feel like another kind of memory, a bedroom that could have hosted early Basecamp's millions in MRR. xAI putting 100,000 GPUs in tents rather than renting from a competitor is the same thesis at pioneer scale. Own for the long duration. Renting for years is the confusion no other industry would make.

DHH does not think Elon Musk primarily *finds* great people. He attracts them with missions stated clearly enough — Mars, electric transport, a networked Earth — that the best show up, haters and lovers in equal Kathy-Sierra measure. The hiring superpower is being able to evaluate the work: smell who is full of it, bet on who gets you to Mars, tell an engineer it can be simpler and be right. 37signals hires on a cover letter (remote work is writing) and a programming test that simulates real work, not on school or pedigree. He has been reversed both ways: perfect CVs with unhireable code, uncertain candidates he wants on the team tomorrow. Steve Jobs had a version of that insult that sparked better work. Apple, he argues, lost the asshole who would say it was not good enough, and gained a logistics genius. The bill came due later: Vision Pro, faltering software quality, a board whose average age he puts around 75, an executive team over 60, and a company behind on AI. He is 45 and already has to force himself to be a beginner again.

He did not always hate Apple. Unix under the Mac rescued him from 1990s Microsoft. He converted half his Copenhagen Business School class with a white iBook, shot a developer ad with Jason in 2004 or 2005, and stood five hours in 2007 for a first-generation iPhone while staff clapped them out of the store. The rot, in his telling, was the App Store toll booth: 30 percent of other people's businesses, plus permission, on the computer that became the most important computer. Basecamp had lived with companion-app rules for a decade. HEY, built for two years as a full-stack Gmail competitor — client *and* server, because you cannot move email if you do not own the mailboxes — got its iOS app approved on a Friday, launched on Monday, and then had an update blocked. Use in-app purchase or leave. It was not only the cut. It was losing the customer relationship, refunds, discounts, and the ability to take a subscription to Android.

They said no, during WWDC week, and made enough noise that Apple took a truce: no 30 percent, stay in the store, build a dummy account so the app "did something" — a rule he says Phil Schiller invented under press pressure. Suing was quoted at $10 million and five to seven years; Epic later revealed well over $100 million. Tim Sweeney and Mark Rein, he says, inflicted the only serious wound in this campaign, won one count out of eleven (the right to link out), and then watched Apple's response anger a judge into a contempt referral. That link-out is how 37signals planned to let HEY customers pay on the open web and carry the subscription across devices. "I will burn this business down before I hand over 30 percent of it to Apple" is the temperature he tries not to live in, and also the temperature he thinks founder-led companies still need. Apple remains special — M chips, taste — which is why the disappointment is worse.

Fatherhood was not on his list in his late 20s. His wife Jamie wanted marriage and three children; he said whoa, then all right, and now calls the alternative an existential regret. The scale of satisfaction, he says, was not one to ten. It was one to a hundred, and he had been playing at the bottom. He would trade the software, the companies, and the race cars in a heartbeat, a cliché he insists is still true because almost no parent he knows ranks otherwise. The West, in his view, is running a *Children of Men* experiment by downgrading marriage and parenthood, and people who chose the traditional path have an obligation to talk it up against a culture fluent in the costs. Kids did not cut his hours — still about 40 a week, the thesis of *It Doesn't Have to Be Crazy at Work* — but they made 9-to-5 strict: *Fortnite* with the boys, school run, dinner, reading, done. Urgency kills fluff. He has a thousand hours in *Fortnite*, all of it with his kids.

He did not get a driver's license until 25, because Danish car taxes are over 200 percent. A friend took him to Chicago Autobahn Country Club; 20 laps in a 1990s Mazda single-seater produced the same pseudo-religious click Ruby had. The intoxicating bit is the edge of adhesion: a few percent of slip, a tiny move from the wall, something real on the line. Programming flow is a lottery; a race car almost guarantees it, because 100 percent of his brain is required. He raced every weekend for four years, reached Le Mans in 2012, and later won the 24 Hours in class. Gentleman drivers can match pros when everything is right, not every time. He has wrecked cars, sprained a hand in a Formula 3.5 a month before a winning Le Mans, watched co-driver Pietro Fittipaldi break both legs at Eau Rouge, and finished second at Le Mans in a year a fellow Dane died in the same car against a tree at 90 g. He still loads a corner-by-corner program from pro traces and tries to become a robot for two and a half hours. Obsession is what Senna and Verstappen have; he does not claim it. His desert-island car is a Pagani Zonda HH from 2010, terrible as a race car, perfect as a road car — the same Costco-Pagani distinction he applied to programming languages.

The programming setup of the last year is Linux with Omakub, a Lofree Flow84, a Logitech MX Master 3, Apple's 6K XDR still on the desk, one screen, Neovim plus LazyVim after a three-day curse leaving TextMate. Linux, for a web developer, is closer to production and can look as personal as the Amiga days. Beginners, he says, should start with Ruby, then JavaScript because the web is the great business platform, then Go for the boring-fast pieces like HTTP proxies — not for business logic. He rejects the identity of software engineer: he is a software writer. Open source, finally, is not a democracy and not a shop. He put a "fuck you" slide up at an early Rails conference when someone demanded Oracle support. Users receive gifts; they do not buy a vendor. Build for yourself, MIT-short, or you will burn out hoping strangers pay you for charity. WordPress is a marvel; Matt Mullenweg's fight with WP Engine, in DHH's October 2024 post, still violates the rule that you do not give a gift and later invoice the people who succeeded with it. He has no dog in that fight except the license. Rails must remain something a company can trust in year seven.

Nirvana, for him, is an empty calendar. Researchers asking the rich how much would feel like enough always heard "twice what I have"; he stopped doubling. Money solves a layer of anxiety and creates another, and never touches the rest — Jim Carrey's wish that everyone get rich enough to learn it was not the answer. Coco Chanel: the best things in life are free, the second best very expensive. Order the list correctly. Asked what gives him hope, he chooses optimism because he does not know either way, and because we suffer more in imagination than in reality. The atom bomb may have prevented World War III. Experts in the pandemic got a lot wrong. "Nobody knows anything" is the refrain he wants to keep. Lex quotes *Rework* on the way out: what you do is what matters, not what you think or say or plan. Inspiration, DHH adds, is perishable. If you stop to write the plan, you may have already lost the thing that would have made you start.