The Silicon ValleyDecoder

Dmitri Dolgov runs Waymo, whose vehicles have completed more than 20 million fully autonomous trips and driven over 200 million fully autonomous miles. He told the room something most people outside tech never hear: by demo standards, self-driving was “solved” back in 2010. The cars could handle freeways. They could handle city streets. The demo worked.

Then it took the better part of a decade and a half to turn that demo into something the public could actually ride. His exact framing: a working demo is 1 percent at best of the work required to build a real product. Getting to the first 90 percent feels incredible. You think you are almost done. The last stretch is where the real company gets built, because physical AI has to be reliable in the real world, not impressive on a stage. Waymo now runs constant evals on its driver, and Dolgov calls them a competitive advantage. The result of all that unglamorous work is a driver that is 17 times safer than humans.

Girl, here is what that actually means. The flashy thing you see on launch day, the demo, the trailer, the viral clip, is the easy 1 percent. The other 99 percent is making it work every single time, for every kind of customer, on the days nothing goes right. That gap between “it worked once” and “it works always” is where most companies quietly die, and it is where the winners earn their money.

This one idea explains so much of the AI economy. It is why a company can raise millions on a demo and still fail. It is why Waymo’s 15 year grind created something competitors cannot copy overnight. And it is a filter you can use on every AI product, pitch, and headline you see from now on: is this a demo, or is this a product?

The demo versus product gap is a career opportunity hiding in plain sight. The 99 percent is operations, quality, trust, customer experience, and testing. That work decides whether an AI company survives. You do not need to train the model to be essential to the company; you need to own a piece of that 99 percent. When you evaluate a startup as an employee, a customer, or an investor, ask the Dolgov question first.

The Photoroom founders opened with the move that made their company: they started doing both video and photo. They cut video and focused entirely on photo, and grew 10 times. Then they narrowed again, to e-commerce photo specifically, and grew 10 times again. Their phrase for it: do not confuse depth with a lack of ambition. Being deep is how you achieve ambition. Today they count 300 million downloads and 20 million active users, with customers like Amazon, Uber, and DoorDash.

Paul Copplestone did the same thing with Supabase. Instead of marketing a database platform to the whole world, he targeted YC startups from day one: a narrow, specific, hungry customer. Now over 60 percent of every YC batch uses Supabase, and the company recently raised about 500 million dollars at a valuation above 10 billion.

Narrowing your focus feels like shrinking your dream. These founders grew 10 times, twice, by cutting things. The trick is holding both at once: a starting point small enough to dominate, and a vision big enough to matter. You win the small room first, and the small room opens the big one.

“Who is it for?” is the question that kills vague ideas and feeds real ones. Everyone is not a customer. E-commerce sellers who need product photos is a customer. YC startups that need a database this week is a customer. Specific customers can be found, talked to, and won.

Women get advised into smallness all the time, so let me be precise: narrow is not small. Narrow is a strategy; small is a ceiling. Pick the niche you know better than anyone, the one you have lived, worked, or shopped inside, and treat it as your entry point to something enormous. Depth first, then the zeroes.

Alexandr Wang, who started Scale at 19 and now leads Meta’s superintelligence lab, named this directly: conviction before consensus. By the time everyone agrees an opportunity is real, it is mostly gone. Blake Scholl lived it: in 2015, Jeff Bezos passed on Boom’s seed round. Scholl kept building, and that same year Bezos wrote in his Amazon shareholder letter about the nature of missed opportunities. Jensen Huang lived a harsher version: in 1995 NVIDIA realized it had bet on exactly the wrong chip architecture, nearly died, corrected course, and went public four years later at 300 million dollars. The company is now worth, in his words, north of a trillion.

Everyone agreeing with you is not the green light. In this world it is often the sign you are late. The people who built the biggest things were told no by the smartest people in the room, sometimes by Jeff Bezos personally, and kept going because they had done the work to trust their own judgment.

Notice what conviction is not: it is not stubbornness or vibes. Wang pivoted to find the right idea. Huang admitted the 1995 bet was wrong and changed course fast. Conviction means trusting your research over the crowd’s comfort, while staying brutally honest about the facts.

Women are socialized to gather permission before acting, and the market quietly punishes that habit, because permission arrives after the opportunity leaves. Build your conviction the way these founders did: through evidence, customers, and receipts. Then let a no from an important person be information, not a verdict. Bezos said no to a supersonic jet that flew anyway.

James Hawkins of PostHog said the thing nobody tells beginners: with serious investors, it is all upside. You cannot win on downside. The investors worth having focus on one question, what if it all works, because they know 99.99 percent of a startup’s upside has not happened yet. A YC partner in the same conversation showed the math underneath: YC has invested in about 7,000 companies, and roughly 90 percent of its returns come from five of them. An investor would rather take a 1 percent chance at a 10 billion or 100 billion dollar company than a sure thing that tops out at 50 or 100 million.

The Photoroom founders turned that math into a habit: when you set a target, add a zero. They practiced saying the word billion out loud until it stopped feeling ridiculous.

This sounds backwards, so stay with me. A modest, sensible plan is actually harder to fund than a huge one, because the investor’s math only works if some bets are enormous. Playing it safe does not read as responsible in that room. It reads as capped.

This is the decoder ring for investor behavior. It explains why funding goes to moonshots, why “realistic” pitches die, and why a founder must be able to answer what happens if everything works. It also tells you when venture money is the wrong tool: a great business built to reach 5 million a year should probably never take it.

Practice the Photoroom exercise on yourself. Take your current goal, add a zero, and sit with the version of the plan that would get you there. Understatement is expensive in rooms that only pay for upside. And if you invest, you now know what the other side of the table is doing: hunting for the five in 7,000.

Blake Scholl built a supersonic jet with a company of roughly 50 people, in an industry where the last entrepreneur-founded commercial airliner company was Douglas Aircraft in 1921. Along the way his team helped change US law on supersonic flight over land. Sam Altman showed the same force in the data: solo founders have jumped from 5 to 10 percent of YC applications to 15 to 25 percent. And a YC partner explained why: an engineer using today’s best AI is massively more productive than one without it, and the cost of that intelligence keeps dropping about 10 times a year.

The size of the team no longer sets the size of the dream. Work that used to demand a building full of specialists can now be attempted by a handful of committed people, and sometimes one. The bottleneck has moved from headcount to judgment: knowing what to build, for whom, and why.

Every barrier that used to justify waiting, no team, no technical co-founder, no budget for engineers, is weakening at the same time. That does not make building easy. It makes it accessible, which is a different and more dangerous word, because it removes the excuses.

The expensive gatekeepers, the technical hires, the agencies, the developer quotes that used to price women out of their own ideas, matter less every year. If a frustration you know intimately could be a product, the distance between you and a working first version has never been shorter. Judgment, taste, and customer knowledge are the scarce assets now, and nobody needed permission to build those.

Paul Copplestone, whose company is now valued above 10 billion dollars, said the thing that keeps a founder grounded is knowing that money will not solve your problems. Alexandre Lebrun, who raised about 1.2 billion dollars for AMI in what was described as the biggest ever seed round in Europe, went further: the real cost of that money is not the dilution. It is the expectations. Raise a billion and show nothing for two years, and it becomes very hard to survive. Olivier Pomel of Datadog added the public market version: his company’s stock once dropped around 65 percent while the business itself stayed sound. Surviving that takes a company built on customers, not hype.

Funding headlines measure fuel, not the engine. Money amplifies whatever a company already is: a strong one goes faster, a weak one crashes at higher speed with more people watching. The founders holding the biggest checks are the ones warning you loudest.

Read funding news differently from now on. A giant raise is not proof a company works. It is proof investors believe it might, and it starts a clock. The durable signals live elsewhere: customers who stay, revenue that repeats, a product people would miss.

This applies to your money too. Windfalls, salaries, and raises amplify existing habits; they do not create discipline. And when you evaluate companies, as an investor or a potential employee, look past the raise to the engine. The question is never how much they raised. It is what happens when the money meets the company.

Blake Scholl started Boom because he could not accept that passenger flight had stopped getting faster after 1969, the year we landed on the moon and flew Concorde. He closed his talk with a charge to the room: go build something you love, and leave this planet better than you found it. Jeff Dean, after nearly three decades of breakthroughs, ended on the same note: build something that truly matters. And Patrick Collison distilled the entire startup playbook into one line: build something people truly need.

After all the numbers, the most senior people on that stage kept landing on the same soft-sounding advice, and it is not soft at all. It is a filter. Building anything real is brutally hard, and the only ideas worth that price are the ones you would be proud to have spent years on, that solve something people genuinely need.

Trend chasing has a short shelf life; the trend moves and the motivation dies. The founders who survive the demo-to-product grind, the rejected seed rounds, and the 65 percent drawdowns are the ones attached to the problem, not the hype cycle. Love of the problem is a durability strategy.

You have spent your whole life noticing what is broken: products that ignore you, services that overcharge you, industries that were not designed with you in the room. That noticing is a founder’s raw material. The question this guide leaves you with is not whether you are qualified to build. It is which broken thing you refuse to leave the way you found it.