ThePrimeagen Would Rather Redesign the Room Than Trust Willpower

Lex Fridman introduces Michael Paulson as ThePrimeagen: a programmer who has made millions of people want to build software and have fun doing it. The first question is not about Netflix or Neovim. It is about joy. Paulson traces his to a data-structures lecture at Montana State University, where Ray Babcock put a linked-list node on the board. The class was Java. The node had a private member of type Node. For the first time he saw a structure that was not a list of ten items. It could hop through memory forever.

Fridman answers with his own version of that explosion: Lisp, Othello, chess engines, and Conway's Game of Life, where each cell only knows its neighbors and the whole board still grows infinite complexity. Paulson's path to recursion was less romantic. Deitel and Deitel told him every budding developer solves the Tower of Hanoi. He could not. Factorial and Fibonacci looked like complicated for-loops. Then a lab handed him a maze in a 2D array, and going up, right, down, and left with a visited grid made recursion click. He had understood it the whole time. He had never had a problem that needed it.

The worst part of programming, he says, is the opposite of that maze: knowing everything. At a company called Schedulicity the days were map-an-object-from-the-database-onto-a-page. No unknown unknowns. He identifies as a tools engineer more than as a front-end or back-end specialist. At jobs he would start on the UI, then get asked for a library, then a build system. At Netflix he spent most of ten years building things other developers needed, including the WebSocket framing for TVs. Developers, unlike most users, usually want the thing they say they want.

Fridman jokes that Paulson mentions Netflix almost as often as Neovim, then asks for the life from the beginning. Paulson starts at four or five, the first time he was exposed to pornography. From then until his early twenties, he says, there was not a day he did not think about it. At seven his father died. Years later, on a trip to China, a rule said to use the word "Dad" instead of "God." Saying it after seventeen years felt like the strangest phrase of his life.

Computers came early, and so did programming. A game called Graal Online, Zelda-like with a level editor and a C-like language, is how he learned: stare at symbols until they make something happen. Socially he was bad. High school was brutal. After his father died, an older brother got into drugs and exposed him to marijuana at eight. From fifteen to eighteen there was a lot of drinking. Around graduation he attempted suicide, went to the hospital, and came out feeling as if something had broken. The social awkwardness died. So did everything else. LSD, mushrooms, and meth followed him into college.

He never really knew what a dad was, which he now reads as a longing that sent him hunting for approval. His brother, older at the time of the death, fell into addiction and never got back out. Paulson thinks seven was, if you have to lose a father, almost the least-bad age: memories stay fuzzy. The hunger for a loving figure did not. He looked for it in sex, in drugs, in being accepted. He did not think that at the time. He thinks it now.

Advice for kids who feel like outsiders is blunt: the people around you in high school will be some of the least consequential people in your life. While you are in it, they are the whole world. Fridman calls it a tutorial level, a chance to take risks before real consequences begin. Paulson will not condone punching anyone, but he agrees that physical confidence and the reminder that the moment is short both help. He also thinks it is harder now. His generation had a qualification of cool. This one has a quantification: follower counts as a public score of how uncool you are.

Porn, he says, was the hardest addiction to quit, harder than cigarettes, because it has almost no social stigma. Tweet that you quit meth and nobody argues. Tweet that you quit porn and friends treat it as optional. It is private, short, and compatible with going out to lunch afterward. The damage he names is not only dopamine. It taught him to use another person as a means. He dated through that lens. People became commodities. Even if every statistic said it was healthy, he would still call it bad for the soul.

Fridman worries that making sex the primary objective devalues intimacy, the way the "manosphere" can. Paulson is more absolute: there is no universe in which meaningless sex is a more meaningful life than fifteen years with his wife. Trust, he argues, is proportional to the depth of a relationship, and vulnerability is a component of trust. Finding her was the second most impactful event of his life. The first was that night with God.

The conversion was not a staged altar call. He was in South Hedges at Montana State, at night, wrapping all his worth in getting laid, and he felt what he describes as the dramatic presence of God. The image he uses is Frodo on a ridge: fall either way. He chose to change, went to sleep, and woke up the same person. That evening he looked at porn and felt shame for the first time in his life. Pot produced the same shame. Hurting someone did too. The gift, as he puts it, was a conscience. After that he did not have a choice.

He started trying in school. High school GPA: 2.14. A teacher once wrote that he was the worst student she had ever had. He failed pre-calculus, then failed it again, then got a C. He used to be the smart math kid. Now he was failing. A year and a half into grinding at a math learning center, something clicked and he became, almost suddenly, the best person in the room. He still cannot say whether the click took weeks or months. It was preceded by a huge amount of difficulty, and then learning itself became desirable.

Quitting porn was still the worst of the addictions. The move that finally worked was realizing he was taking something from a future wife he had not met. Replacing porn with a girlfriend would not have stopped the problem. The distance from head to heart, he says, is the greatest in the universe. He quotes a depressed writer's image of antidepressants as a drug that lets you see rain while the crop still dies. Porn was a search for meaning that could not provide it. That is why it is addicting. By about 22 the outward addictions were done. He met his wife not long after, watched *The Social Network*, started a PHP startup that night, and programmed 12, 24, 36 hours at a stretch.

Fridman asks him to reverse-engineer the wall. Paulson speed-runs it: pre-calculus three times, then Calculus 1 over the summer, a two-hour final finished in 30 minutes, highest score in the school. He was the only person out of 400 to finish the differential-equations final. The method was not insight. It was obligation. He would not graduate otherwise. Class, then the math center, then the odd-numbered problems in the back of the book until inverse trig substitution lived in his head like a times table. He is bad at going from abstract to concrete. He is good once he has done a thing literally.

He hates the phrase "work smarter, not harder." The hidden suggestion is that you already know what smarter is, so stop trying so hard. He has never experienced that. He learns what smart looks like by working very hard with no shortcuts, then combining the two. Fridman, who learned guitar, piano, and jiu-jitsu the same expensive way, agrees: great engineers forget the hours once the work feels easy, then they give the stupid advice of taking the easy path. Paulson's addendum is a T-shirt: work hard, get smart. He also hates "the journey is better than the destination," which, taken literally, means you are always walking toward something worse.

A mid-podcast insert thanks his mother, who worked 18-hour days after his father died and became the internal reason he wanted to get his life in gear. The path to Netflix was not a leap from the math center. He spent a couple of years on a failed SMS-marketing startup, Via Text, which he now sees as five years too early, the way his father's early-90s prepaid long-distance idea was too early. An unhinged downstairs neighbor who threatened to kill them over a dropped AA battery forced a real job. Schedulicity was boring. WebFilings demanded 60-hour weeks and complex UI state machines. A Netflix manager, Jeff Wagner, wrote a personal note about RxJS. Paulson walked in after five years of 80-hour weeks intending only to show what he already knew. He called his wife mid-interview: they were going to have to decide whether to move to California. She was 36 weeks pregnant. They knew no one. Netflix's culture deck still said hire fast, fire fast.

He joined the TV team in 2013 and was handed a PlayStation 3. The middle layer between backend and UI was Groovy, a JVM language almost nobody on a team of forty understood. Someone had hacked scripts together until they ran. His first experiment, `observable.sleep(1)`, put a thread to sleep that nothing could wake. He locked staging for the afternoon. He gravitated toward the mess because he is good at print debugging. He rewrote a TV data structure from a list of lists of movies into a list of lists of recommendation objects, so the same title could carry different evidence in different rows. He learned Groovy by reading the language reference for a day.

The skill came from a student contract writing robot drivers for the government: missing manuals, Ethernet, motors, and the only output channel was print. At Netflix the machines lived on AWS he could not log into, so the same muscle applied. A colleague's goofy Falcor request-deduper looked wrong, so Paulson deleted it and rewrote it. By the end he had reproduced the same code in his own style. The lesson he kept: solved problems are already in the codebase. Starting from scratch is how you learn a concept, not how you replace a working library. Falcor later produced a named vulnerability, Repulsive Grizzly: a compact path request that asked the JVM to materialize billions of objects. One request killed one machine. Fifty requests could freeze the UI across web, TV, and mobile. Video might still stream. You could not get to a video. It had been in production for years, so there was no rollback to last week, only to 2011.

He has broken production for stupider reasons: PHP `static` that survived across requests and mixed country data; a Lady Gaga premiere billboard that froze at zero; an infinite loop on "add to my list." Pieter Levels shipping PHP, jQuery, and SQL with almost no process is, in Paulson's view, correct for a solo app and a disaster at Netflix scale. The Black Mirror "pig" episode of season one had to stop being the first thing people saw when Netflix launched the show. Reversing season order pulled in about twenty engineers across metadata, services, web, mobile, and TV. He was named captain, borrowed people from each team, and they shipped in roughly three weeks. Small features, at that size of company, are not small.

Streaming started as Extra Life, a charity that sends money to kids' hospitals with no overhead. A friend he calls the greatest engineer he has met, Guy Cirino, dragged him in. He learned to stream on Fortnite, got affiliated, then one day programmed in Vim on stream and thirty people showed up. For years the schedule was Netflix, dinner, Lord of the Rings with the kids, then three nights a week until 2 a.m. Pirate Software handed him a challenge coin at the Streamer Awards and told him to go full-time. Leaving Netflix was emotional turmoil: a safe, fun job versus a public that can flip. He told himself he was getting older, and this might be the last crazy thing he would get to try. April of the previous year, he jumped. He was not yet a year in when they recorded.

The name is the final boss of Turok II on Nintendo 64, a brutally hard game he beat as a fifth-grader after a month of ten-hour days. As an adult he used Michael Paulson on the internet until around 2017, when tech Twitter felt like a pretentious, cancel-prone measuring contest. He changed the handle back to the video-game name and made the stream a project: destroy the mentality that smart people design for dummies. Newcomers should be allowed to ask dumb questions and reinvent the wheel. Arnold Schwarzenegger's unpronounceable name became the brand; ThePrimeagen did something similar in miniature. The mustache began as No-Shave November and turned into a lifestyle his wife agreed to tolerate. His editor, Flip, took half of YouTube revenue for years when that number was roughly zero. The work is still lonely. He misses watching other people solve problems, even solutions he hated.

Learning a language, he says, has two honest methods and no shortcuts past building. The first is to Google your way through a main function, ifs, and modules, which gets you shipping but imports every previous language's habits. The second, inspired by Mitchell Hashimoto, is to read the whole language reference, eight to ten hours for Zig, so the features are indexed before you write. Juniors should not start there. They should grow complexity by hand. AI as a tutor can ask questions in your own voice, which he finds powerful, and it can also produce learned helplessness: a new error, a pasted fix, no knowledge. He worries about a generation of juniors who remain juniors because unique work still needs elbow grease.

Asked for the best language in 2025, he refuses the Stack Overflow survey as biased toward new users and people who love answering questions. The first language, if you know nothing, should be JavaScript, a sentence he knows will make people angry. You can draw a box, move a box, write a backend, hit a database, and put it on a canvas. Python wins for data and ML but cannot give you a front end. HTML and CSS are not programming languages by his practical test: if you cannot write a function, an if, and a loop, he does not count it. Bash is a superpower until you need an array. FFmpeg, he and Fridman agree, is one of the most consequential libraries alive, under-loved, and sitting under OBS and almost every media tool.

A sensible diet is one dynamic language, one strict one (Go, Rust, C#), and then whatever you are curious about. His curiosity list is Swift, Elixir, OCaml, and Odin. Go is his mainstay for backends with complicated logic, painful on heterogeneous lists. Side languages for the year were Jai and Zig. Zig sits between C's expressiveness and Rust's safety: optionals stop some null dereferences without making every pointer a ceremony. What he loves in Rust is tying heap cleanup to stack scope, and making mutability explicit. Metaprogramming in Zig is just the language plus `comptime`. An epic stream project spawned ASCII Doom, piped it to a website at about 70 milliseconds of lag, let Twitch chat vote on WASD, and compressed the frames from gigabytes to kilobytes with quad trees, run-length encoding, and luminosity-to-glyph tricks inspired by FFmpeg.

He learns the most, he says, around 5,000 to 10,000 lines, when early design choices either weigh you down or free you. Fridman's assert-everywhere habit, John Carmack, and Joran of TigerBeetle pushed him to put two asserts in every function and run deterministic simulation. He found bugs in a reverse proxy and a game that he would never have found by clicking around. Twitch chat, compared with YouTube, shares a meme bloodstream; YouTube chat is more earnest and can "swim" to old timestamps, which makes it feel disjointed. He does not think the editor makes the programmer. He does think it says something about character if you will spend 50,000 hours in a tool and never learn it.

Fridman asks how he flourishes with ADHD. The unglamorous answer is maturity: reread the page eighteen times in college rather than agree with the voice that says you cannot. Emotional resilience, built over years, let him stretch programming from a short burst into 10, 12, 15 hours, which he now loves the way some people love an instrument. The rest of life is another story. Scheduling a dentist, packing kids' meals, the administrative snowball: historically terrible. He will not claim he defeated ADHD alone. His wife clears a lot of the debris. He calls her a cheat code. The places he can control, he has engineered. The places he cannot, she covers.

Focus in programming, for him, is the absence of side quests. Early on every step was a question: double buffering, writing to the screen, why is it blinking. Once the instrument is learned he can write a thousand lines on an airplane without looking anything up. That is also why he wants people to type without looking and to learn the editor until the motion is a reflex: looking down at a key you already know is still a tax on working memory. Twitter still pulls. Plugin rabbit holes still pull. His rule is that he edits his Neovim config once a year unless something is on fire. Last month he spent a hundred hours going zero-to-sixty on a new machine. Then he declared the breakpoint and stopped. A perfect editor is a pursuit that never ends.

The setup is a Kinesis Advantage that he says saved his career, one monitor, Linux with i3, a yoga ball, and StarCraft-style hotkeys: Alt plus a number to a named workspace. Tmux sessions spawn Vim, a spare terminal, and long-running processes. Ctrl-F fuzzy-finds any project folder. Harpoon pins up to four files to single keystrokes, because a feature usually lives in three files and search fatigue is, in his phrase, a massive fail. He was a Vim hater, Saul before Paul. Motions are a language for editing, not for typing: `dap` deletes a paragraph, `ci"` changes inside quotes. The dividends take a long time. If you will not endure the pain, they are worse than a mouse. Harpoon 2 is done on a branch; he keeps master on v1 because he is allergic to being an active maintainer after Falcor crushed that part of his soul. He builds tools for himself. Three months of navigation pain produced four power-finger jumps. That is the whole product.

His general view of AI is simple: it is nice when the work is predictable, and not nice when it is not. It is strong at TypeScript and weak at Zig. Nat Friedman saw a Twitch clip of him begging GitHub for Copilot and sent early access. The first correct if-statement blew his mind. The more he used it, the more bugs he introduced, and the realization was ugly: he was not copiloting well, he was autopiloting. Reading code and writing code are different skill levels. He stopped around six months before the recording. Fridman uses Copilot and Cursor as pair-programming against loneliness, and as a way to learn the skill of editing model output. Paulson does not deny that joy. He felt less engaged, more willing to let things slide.

A lot of his functions, he says, are chess moves: they exist because 2,000 lines later he will need the next activity. When generation speeds that up, the chess falls apart. He calls that a skill issue in the honest sense, and he has been meaning to spend two weeks fully inside Cursor. Fridman argues that avoiding the tools is falling behind, because prompt work builds empathy for a system with no common sense and no long-term memory. Paulson pushes back: if the interface is going to change so much that today's prompt library is punch-card AI, have the holdouts actually fallen behind? He already uses models heavily off the coding path: iterating Randall-the-engineering-manager prompts on Twitch chat, pitting Claude 3.5 against OpenAI in a Zig tower-defense game, giving chat a voice through ElevenLabs. The prompting skill is real. It has not yet paid off in his editor.

The gripe is not capability. It is culture. Someone replied to Pieter Levels that nobody in 2025 should be acquiring hard skills. Paulson thinks that is damning advice for the young. Models get code wrong, especially as systems grow, and they slosh between bugs. If you are not the driver, your progress is bound to the LLM. He will not predict the year programming becomes unnecessary. He will say every computing jump so far created more jobs, that natural language is still too ambiguous to be the program, and that a less-ambiguous "pseudo-lang" for models would just be another higher-order language. More people able to program means more software to maintain. CRUD table displays were already Lego. The people who can hold a bigger project, up and down the stack, get more valuable if models can take the intern tickets.

Devin's pitch, going off to explore a repo, commit, and knock down backlog tickets, is a thing he wants badly. The people who run it, he stresses, are extremely nice. In practice the more he narrows a task, the better it does, and he still rewrites the result and wishes he had just learned Tailwind to the nth degree. Debugging is where humans still look special: logical bugs, rare edges, fog. He likes the curl maintainer's line that the I in LLM stands for intelligence, written after a flood of hallucinated security reports. He names a coming attack he calls denial of attention: fake bug reports and hounded maintainers, in the spirit of the Polyfill supply-chain mess. Stack Overflow's survey showing 25 percent with no plan to try AI surprises him. Trusting model output highly, in his view, is something only the very green would do.

Advice for someone in their early twenties is not a language list. Get the reps, find the flavor that makes you come back every day, whether that is agents or Zig or JavaScript, and do not get lost in the sauce: health, relationships, and morals are easy to shortcut for a startup that will not love you back. Would he still tell a young person to become a programmer while models improve? He wants his own kids, a decade younger than that hypothetical twenty-year-old, to learn. Maybe they throw the skills away in twenty years. He bets an entirely new field will exist, the way social networks would have sounded like nonsense in the 1970s. A first job can make you hate programming; his C# internship at Total Information Management System sent him into mechanical engineering for a semester. Stay if the work is alive. Leave if you have gone mindless. Money sometimes shows up when you do not strive for it.

Kids, he says, are not the career trade-off people pretend. You do not know your capacity for love until you have them. Dogs recede. A medical hypothetical in which a parent dies so a child lives is, for him, not a question. Fridman notes that some of the most productive people he knows have children. Paulson says it changed how he uses time. On Microsoft he is wary of EEE by another name: VS Code, Copilot, GitHub, and CI as a trap whose switching cost eventually becomes infinite. He still grants that VS Code is a good product. Neovim, with Justin M. Keyes as benevolent dictator, cannot cash that trap. Terminal.shop, the SSH-only coffee company he built with Teej, Dax, Adam, Liz, and David Hill, exists because he did not want to shill a tool that could wreck someone's startup. If you cannot SSH, he jokes, you probably should not buy the coffee, though you can learn.

Fridman comes back to God near the end. Paulson grew up a functional atheist, church a handful of times, no strong family tie. The night at 19 still looks like the turning point of the entire life. Without it he does not think he would have finished college, found his wife, had his kids, or learned how to value people. He quotes a verse he cannot place: we love because he first loved us. For the first time he did not have to live up to a standard. The standard had been paid for. He did not have to be the cool guy or win a sexual conquest for validation. It was, he says, exceptionally liberating. Who God is, pushed too hard, is how you fall into heresy. What God did, for him, was make happiness and forgiveness possible.

He wrote down the names of people from high school, held the list, and chose to forgive them. Hate is sticky for a lifetime. He does not think you get rid of it without forgiveness, and he would not have known forgiveness as a method if he had not first been forgiven. Love, narrowed past eros, is paternal, friendship, and agape, each applied in a different place. He does not see a good world without that mixture as a foundation. If forgiveness pays off hate, love has to be the motive. Fridman calls loneliness and longing the tragic half of that picture, and love the beautiful half.

A Reddit note asks Fridman to thank Paulson for everlasting positivity and to give him a hug. Fridman does the thanks without the hug. Listeners going through a lot of shit, he says, find that even when Paulson is building the stupidest things, the positivity radiates and makes them want to be decent and to build. Paulson is surprised the kindness came from Reddit, then turns it around: the best part of the public-facing work is the chance to improve a life, the way someone might have learned Swift from a Chris Lattner interview on this same show. Fridman closes with Paulo Coelho: when we strive to become better than we are, everything around us becomes better too.