(1) PROBLEM FIRST — not technology, not solution, not market size. A specific, painful problem experienced by specific, nameable people who are currently paying time or money to solve it badly. (2) DO THINGS THAT DON'T SCALE — manual, personal, high-touch actions for early users. Personally deliver. Personally onboard. Personally support. Learn from every interaction. You cannot learn at scale what you refuse to learn manually. (3) REAL METRICS — weekly retention (do they come back?), MRR (do they pay?), default alive/dead (at current growth rate, do you reach profitability before running out of money?). Not signups, not page views, not followers. (4) TALK TO USERS — every week. Not surveys — conversations. What surprised you? What did they say that you did not want to hear? What changed because of it? (5) ONE THING — what does your product do 10× better than the alternative? One sentence. If you need two sentences: you are not focused enough.
Structured reflection tool that forces the LLM to think like Paul Graham — Y Combinator co-founder, essayist, and the voice behind "Do Things That Don't Scale" (2013). YC has funded 5,000+ companies with $1T+ in combined valuation (Airbnb, Stripe, Dropbox, Coinbase). The YC method is counterintuitive: start with a problem, not a solution. Do manual, unscalable things for your first users. Measure retention, not signups. Talk to users every week. Focus on one thing done 10× better, not 10 things done adequately. Catches Solution Seeking (building technology before validating a problem — a potter builds an automated glaze-mixing machine. 6 months of engineering. $15,000 in materials. It mixes 47 glaze recipes perfectly. Shows it to 20 potters: "How do you currently mix glazes?" "By hand. It takes 5 minutes. It is part of the craft. I enjoy it." "Would you pay $3,000 for a machine that does it?" "No. Mixing is not my bottleneck. Kiln firing schedules and clay procurement are." The potter built a solution for a problem that does not exist. PG: "The best startup ideas come from problems you personally experience — not solutions you find technically interesting." Rule: name 5 specific people who HAVE this problem, describe their CURRENT solution, and quantify their willingness to pay. Not "people would want this" — "Maria pays $X monthly for Y"), Premature Scaling (hiring, marketing, and expanding before finding product-market fit — a cheese maker creates artisanal goat cheese. 4 flavors. Sells at one farmers' market. Week 3: "We should be in 20 stores!" Hires a sales rep ($4,500/month). Rents commercial kitchen ($2,800/month). Orders branded packaging ($3,200 minimum run). Revenue: $1,200/month. Burn: $10,500/month. Runway: 4 months. The cheese maker has not validated: which flavor sells best, what price point works, whether store customers buy differently than farmers' market customers, or whether the cheese survives 5-day shelf distribution. PG: "Do things that don't scale." The cheese maker should: personally deliver cheese to 3 stores. Stand behind the counter. Watch who buys. Ask why. Learn that the herb flavor outsells plain 4:1. That $8 price point is too high for grocery but perfect for specialty. That customers want 150g portions, not 250g. THEN scale — with validated knowledge), Vanity Metrics (celebrating activity instead of measuring value — a tutoring service: "We have 2,000 followers on Instagram! 450 people downloaded our free study guide! 120 signed up for the newsletter!" Revenue: $0. Paying students: 0. Followers are not customers. Downloads are not revenue. Signups are not retention. The ONLY metrics that matter for a tutoring service: How many students are paying? $[MRR]. How many students come back for a second session? [retention %]. How many students improve their grades? [outcome metric]. PG: "Startups die from vanity. They celebrate signups instead of measuring retention. A startup with 100 users who love you is better than one with 10,000 who signed up and never returned." Rule: weekly retention curve, monthly recurring revenue, default alive or dead calculation), Ivory Tower (building without talking to users — a woodworker designs a modular shelving system. Beautiful CAD drawings. Patent filed. Website built. Launch video produced. 8 months of work. Zero user conversations. Launch day: 14 orders. 9 returns within 30 days. Return reasons: "Too heavy to wall-mount alone" (2 people needed — they live alone). "Modules do not fit standard wall stud spacing" (16 inches, not 18). "No cable management for electronics" (the #1 use case was media shelving). If the woodworker had TALKED to 20 potential buyers before building: "I live alone — can I install this myself?" "My wall studs are 16 inches apart." "I need somewhere to hide cables." These three conversations would have changed the entire design. PG: "Make something people want" — and you discover what they want by ASKING THEM. Rule: how many user conversations THIS WEEK? What surprised you? What changed because of it?), and Feature Bloat (building 20 features instead of being exceptional at one — a food truck: serves burgers, tacos, sushi, pizza, pad thai, acai bowls, and crepes. Menu: 47 items. Kitchen: one person. Wait time: 25 minutes. Quality: mediocre — everything is "okay," nothing is "amazing." Customer review: "Jack of all trades, master of none. The burger was fine. But the truck down the street does ONE thing — smash burgers — and they are incredible." The smash burger truck has 1 menu item with 3 toppings. Wait: 4 minutes. Revenue per hour: 3× the 47-item truck. Because ONE thing done 10× better beats 47 things done adequately. PG: "Do one thing well." Rule: what is the ONE thing your product does 10× better than the alternative? If you cannot answer in one sentence: you are doing too many things). Call once per startup strategy, business plan, or market validation