Early 1970s, suburban Pittsburgh. Paul Graham gets hooked on the idea of computers before he's ever touched one — TV and film versions, electronic brains with limitless power, a near future where you'd just talk to a machine and it would do what you wanted.1
Junior high got him a real one: a gifted-student program let him work on a project of his choosing, and he picked the school's IBM mainframe, used for printing grade reports. Primitive, punch-card programmed, and it felt like a portal to the future anyway.2
He taught himself to program mostly by trial and error, consulting the few books that existed on the subject but learning mainly by doing. The results showed up immediately — like painting, if the code worked, it had a kind of aesthetic rightness to it. Learning this way, without anyone laying out a fixed path, was deeply satisfying. Greene's gloss: this is the essence of being a "hacker."3
Cornell for undergrad, one of the strongest computer science departments in the country — real instruction in programming fundamentals, which cleaned up a lot of the bad habits he'd developed alone.4 Then Harvard for graduate school in AI, chasing what seemed like the field where his childhood dream of talking computers could actually happen.
At Harvard, the confrontation: he wasn't built for academia. He hated writing research papers. The university's approach to programming stripped out the exact thing that had made it fun — discovering through trial and error. He was a hacker at heart.5
He found a fellow hacker, Robert Morris, and together they dug into Lisp — a language that seemed to make you understand something essential about programming itself, built specifically for investigation and discovery rather than for following a predetermined structure.6
Disillusioned with Harvard's computer science department, Graham built his own graduate curriculum by taking a wide range of classes to find what actually interested him — and, to his own surprise, found himself pulled toward art and art history.7
He followed it. Finished his PhD in computer science at Harvard, then enrolled at the Rhode Island School of Design, then a painting program at the Accademia in Florence. Came back to the States broke, determined to paint, paying rent with occasional programming consulting.8
Looking back on this stretch, Graham himself described it — Greene quotes him directly — as having no real design. Like the "cheesy hacks" of his high school years: patching things together, figuring things out through constant trial and error, discovering what worked by doing it. Learned mostly what to avoid: academia, large companies, anything political. What mattered to him was making things, and having live possibilities to move toward, depending on what life presented. If he'd undergone an apprenticeship at all, it was almost accidental.9
Written in Lisp, chosen for the speed with which changes could be made to it. They called the company Viaweb — the first of its kind, pioneering online commerce. Three years later, Yahoo! bought it for $45 million.11
Ten years after that, giving a talk at Harvard about the Viaweb experience, students pushed him to start some kind of consulting operation for young entrepreneurs. He built Y Combinator instead — an apprenticeship system for startup founders, his firm taking a stake in each one, refined the same way everything else in his life had been: by trial and error, as he went. Greene calls it his ultimate hack — something he stumbled onto and improved through the exact process this whole strategy describes. By the time of the book's writing, the company was valued near $500 million.12
Greene situates this historically. Every age produces an apprenticeship model suited to how it actually produces things. The Middle Ages, with capitalism's birth and the need for quality control, produced the rigid guild-apprenticeship terms. The Industrial Revolution made that model largely obsolete but kept the underlying idea alive as self-apprenticeship — developing yourself from inside a field, as Darwin did in biology, matching the era's growing individualism.13
Now, in the computer age, Greene argues the hacker's trial-and-error approach to programming is the most promising available model for apprenticeship generally — not just for programmers.
The model, stated as instruction: learn as many skills as possible, following wherever circumstance leads, but only within the boundary of your deepest interests. Value the process of self-discovery and making things of real quality over following a single pre-set career path. Move by trial and error. This is what your twenties are actually for — you're the one programming this wide-ranging apprenticeship, inside the loose constraints of your own genuine interests.14
The reassurance he attaches to the apparent aimlessness: you're not wandering out of fear of commitment. You're expanding your skill base and your possibilities. At some point, when you're ready to settle, opportunities present themselves, and every accumulated skill becomes usable — you become the one able to combine them in ways unique to you. Rigid, singular career paths chosen young, Greene warns, tend to produce a dead end in your forties. The wide-ranging trial-and-error apprenticeship of your twenties produces the opposite: expanding possibility as you age.15
You're choosing between a structured, credentialed path and a looser, more exploratory one, and the structured path feels safer. Graham's own account undercuts that instinct directly — his most valuable apprenticeship (the hacking) preceded and outlasted his most credentialed one (the Harvard PhD), and the credential didn't end up being what mattered.
You've abandoned a promising direction because it wasn't working and you're worried the abandonment itself was the mistake. Check whether you extracted the "what to avoid" lesson Graham describes — academia, large companies, anything political, in his case. If you can name specifically what you learned to avoid and why, the abandonment was doing its job.
You're evaluating a side interest that has no obvious connection to your main path — Graham's art history detour is the case. Don't require the connection to be visible yet. His painting years fed nothing directly into Viaweb's success, and yet the trial-and-error discipline itself, honed across every direction he'd ever tried, is what let him build something nobody had built before when the actual opportunity arrived.
You're in your twenties and feel behind because your path looks incoherent next to peers who chose one lane early. Greene's argument, backed by Graham's own case, is that the incoherence at this stage is closer to the strategy working than to the strategy failing — provided each move is genuinely following your interest rather than genuine drift with no interest behind it at all.
Single case, and an unusually strong source — Graham's own quoted self-assessment, not just Greene's narration of events, giving this page more direct testimony than most of the book's other cases.
The tension worth naming honestly: Graham's story is a spectacular survivorship case. Viaweb's success depended on a specific, narrow window — being early enough to build the first server-side commerce software before the field crowded, with the right co-founder, sold at the right moment to the right buyer. The trial-and-error method is fully genuine and well-documented in his account, but Greene presents the method as the explanation for the outcome, when the outcome required the method plus a considerable amount of timing that had nothing to do with how deliberately Graham was exploring.
Open question. Greene's model requires enough of a financial floor to survive the wandering — Graham funded his art years partly through consulting work, and never faced genuine destitution. What does trial-and-error apprenticeship look like as a strategy for someone without that floor, where a wrong turn in your twenties carries real, non-recoverable cost rather than merely delayed opportunity?
The claim that a rigid, singular path chosen young produces a dead end by one's forties sits in unexamined tension with Discovering Your Calling's celebration of figures — Curie at four, Bergman's "it chose me" — whose defining feature is that they identified a singular direction extremely early and never meaningfully deviated from it. Greene holds up both the wanderer and the early-committer as models within the same book's argument without reconciling which condition actually predicts the dead end he warns against.
Experimentation — The Active Mode — Greene frames Experimentation as the closing step of a bounded apprenticeship; Graham's entire adult life, on his own account, never exits that mode and instead scales it up into a permanent operating method. The insight the pairing produces: this page is effectively proof that Step Three doesn't have to end where the book's own chapter structure implies it does. Some careers are built entirely on staying in the exploratory phase, converting the "shortest step" of Greene's model into the whole method.
Explore/Exploit Trade-off in a Creative Career — Graham's twenty-year stretch of unstructured hacking, painting, and consulting reads, in explore/exploit terms, as an unusually long explore phase that only converted once — decisively, at Viaweb, and then again at Y Combinator. The insight the pairing supplies is a corrective to how "advance through trial and error" can be misread as endorsing exploration with no conversion point at all. Graham's own account shows the trial-and-error phase running for roughly two decades before either major exploit event; the strategy isn't infinite wandering, it's sustained exploration held open until a genuinely strong signal arrives.
Sharpest implication. The hacker's core discipline — see the result immediately, adjust, try again — is a description of a feedback loop, not a personality type. Which reframes "trial and error" from a temperament some people have and others lack into an environmental condition some fields supply (fast, visible feedback) and others actively suppress (slow, opaque, or political feedback), meaning the strategy's real precondition is choosing domains where the loop can actually run.
Generative questions.