A developer's honest take on the most counterintuitive startup advice ever written
Paul Graham wrote "Before the Startup" in 2014 for a guest lecture at Stanford. It was aimed at college students. It's now 2025, and I just re-read it for the first time in years.
My honest reaction? Most of it aged like fine wine.
Some of it needs a 2025 update.
Here's my breakdown of all six of PG's counterintuitive points — with modern context added for developers and founders building today.
- Startups Are Still Counterintuitive — But Now AI Makes It Worse PG's core warning: "If you trust your instincts starting a startup, you'll make a lot of mistakes."
This was true in 2014. In 2025, it's turbocharged.
Here's the new trap: AI tools are so good at looking like progress that your instincts now feel validated faster than ever. You can spin up a landing page in an hour, generate a pitch deck, get a brand name, write your Terms of Service, and deploy an MVP — all before talking to a single user.
Everything looks right. But nothing has been tested against reality.
The instinct to build first and validate later is older than startups, and AI has made it easier to indulge than ever before.
The 2025 update: Don't trust your instincts about momentum. AI gives you the feeling of velocity. Real progress is still measured in users who come back — not in shipped features.
- You Don't Need to Be a Startup Expert — You Need to Know Your Users This is the point I've seen ignored most aggressively in the last two years.
PG puts it bluntly: Mark Zuckerberg didn't succeed because he understood fundraising. He succeeded because he understood his users better than anyone else did.
In 2025, the equivalent mistake looks like this:
Spending weeks reading about pre-seed valuations
Obsessing over which accelerator to apply to
Building in public before anyone actually uses what you built
Knowing all about safe agreements and cap tables before you have a single paying customer
Meanwhile: when was the last time you sat down with five potential users and just listened?
The 2025 update: With AI handling more of the "how to build" questions, the competitive edge has shifted even more toward "what to build and for whom." User research is now the moat that AI can't shortcut for you.
- Gaming the System Still Doesn't Work — But the Games Have Changed PG's observation: schools train you to find tricks. Startups punish you for looking for tricks.
In 2014, the games were: get a good valuation, rent a cool office, hire fast, look busy.
In 2025, the games are:
Going viral on X or LinkedIn before you have product-market fit
Getting into a prestigious accelerator as proof of legitimacy
Building in public as a substitute for building something useful
Launching on Product Hunt to hit a vanity metric milestone
Getting a big model company to feature your AI wrapper as a growth strategy
None of these are inherently bad. The problem is when they become the goal — when you're optimizing for the appearance of traction instead of actual traction.
PG's answer in 2014 is still the answer in 2025: make something people want. The system you're gaming has changed. The lesson hasn't.
The 2025 update: Distribution loops and virality are real tools. But if you reach for them before you have retention, you're speedrunning the "gradually realize how completely f***ed we are" phase PG describes.
- Startups Are All-Consuming — And Hustle Culture Has Lied to You About This This is where PG's 2014 essay was right in a way almost nobody listened to.
He wrote that starting a startup takes over your life in ways you can't fully imagine beforehand. That it doesn't get easier; the problems just change. That there is a real opportunity cost, especially if you're young.
What happened between 2014 and 2025? We built an entire aesthetic around that cost and called it a virtue.
"I slept 4 hours." "I haven't taken a day off in 8 months." "I missed [life event] for a sprint." These became LinkedIn captions. Aspirational content.
PG's point was never that this sacrifice doesn't happen. It does. His point was: are you sure you want to push that button right now? Because once you do, a lot of other doors close — not forever, but for a long time.
For developers especially: your early 20s are an unusually good time to go deep on weird problems, contribute to open source, explore completely unrelated domains, fail cheaply on side projects. That kind of exploration directly feeds good startup ideas later. Burning that window on a startup that isn't ready yet has a real cost.
The 2025 update: Remote work and AI tooling have made solo building more feasible than ever — which is great. But "easier to build" doesn't mean "lower stakes emotionally and mentally." The all-consuming nature hasn't changed. Be honest with yourself about the timing.
- You Cannot Know in Advance If You're Ready — So Stop Waiting for the Sign PG's most honest point: for nine years, his job was predicting which founders would succeed. And he learned to keep an almost completely open mind, because there was almost no correlation between initial confidence and eventual success.
The swaggering recruits were no more likely to be tough than the quiet ones.
This cuts both ways, and I think developers specifically need to hear the second implication: your uncertainty about whether you're "ready" is not meaningful data. It doesn't mean you're not ready. It means startups are a domain where you only discover your actual capability by attempting it.
The one exception PG notes: if you're genuinely terrified, maybe don't. But if you're merely unsure — that's just Tuesday.
The 2025 update: Imposter syndrome has been discussed so much that it's almost lost its meaning. But the underlying reality is still real. The skills that make someone a good founder — resilience, adaptability, the ability to move through ambiguity — don't show up on any test you've taken so far. You find out by doing.
- Stop Hunting for Startup Ideas. Start Living at the Edge of Something Real. This is the one that most people intellectually agree with and practically ignore.
PG's argument: the way to get good startup ideas is not to try to think of startup ideas. Conscious effort in this direction produces ideas that are bad and plausible-sounding — the worst combination, because they're convincing enough to waste your time on.
Apple, Yahoo, Google, Facebook — all side projects first. None of them set out to be companies.
His prescription for college students (and I'd extend it to developers at any stage): learn deeply about things that matter, work on problems that genuinely interest you, with people you like and respect. Ideas form in that environment without being forced.
In 2025, there's a specific version of this worth naming for developers:
Live at the edge of AI capabilities. Not by building another chatbot wrapper — by actually understanding what these models can and can't do, where they break, what workflows they change, what they make possible that wasn't possible before. The people building the genuinely interesting things right now are the ones who deeply understand the technology's real frontier, not the ones who read about it on Twitter.
The same logic applies to whatever edge you're closest to: embedded systems, biotech, climate infrastructure, developer tooling, fintech in emerging markets. Get to the edge. Work on something real. The ideas will come.
The 2025 update: PG said to "live in the future." In 2025, the future is moving faster than it ever has. The edge keeps shifting. The discipline of staying at it — going deep, staying curious, building things that ought to exist — matters more now, not less.
The One Thing That Genuinely Changed
PG ended with two words of advice: just learn.
I'd keep that advice exactly as written. But I'd add one honest 2025 caveat:
Learning now includes learning with AI — and you have to be careful that that doesn't become a shortcut around actual understanding.
Using Claude or GPT to go faster is fine. Using it as a substitute for actually understanding your users, your domain, your codebase, or your own thinking — that's the trap. The depth of understanding PG is describing doesn't come from prompting. It comes from doing the hard, slow, sometimes boring work of genuinely knowing something.
That part hasn't changed at all.
Summary
PG's Point Still True? 2025 Update
Trust your gut about people, not ideas ✅ Yes AI hype makes bad ideas feel more credible — stay skeptical
Know your users, not startup mechanics ✅ More than ever AI handles "how to build" — user insight is the new moat
Gaming the system doesn't work ✅ Yes The games have changed (virality, launchers, AI PR); the lesson hasn't
Startups are all-consuming ✅ Yes Hustle culture romanticized this; the cost is still real
You can't predict if you're ready ✅ Yes You find out by trying — uncertainty is normal, not a stop sign
Don't hunt for ideas — develop taste ✅ Yes "Live in the future" now means living at the AI/tech frontier
If you haven't read the original essay, it's freely available on Paul Graham's website at paulgraham.com. Worth the 20 minutes.
What's a counterintuitive startup lesson you've learned the hard way? Drop it in the comments.
Top comments (0)