Reps as pattern recognition

Frame from sail-muddy-lessons for what experience actually compounds into. The figure-drawing analogy threads through: dozens of 60-second gesture drawings per session, most of them bad, but the eye sharpens by volume.

The receipts

• Parker Conrad — built Zenefits (YC W13, peaked at $4.5B before he was forced out), started Rippling six weeks later. Says you don't learn much from failure explicitly, but the accumulated pattern recognition makes you faster (YC interview).
• Karri Saarinen — Linear hit PMF in about a year, fourteen months from start to over a thousand paying customers on near-zero marketing spend. He had Kippt (acquired by Coinbase), Coinbase, and Airbnb behind him. Famous for building a Chrome extension at Airbnb that restyled Jira (~100 installs) — that extension foreshadowed Linear.
• Ivan Zhao (Notion) — first attempt didn't work; laid off the team, sublet the SF office, moved to Kyoto with Simon Last and rebuilt from scratch. The "early Notion" everyone references skips this part.
• Nikita Bier (TBH) — the team built 15 failed apps before TBH sold to Facebook. Their first app took a year; their last took two weeks. From his talk: "the most valuable thing you can have is a reliable way to test your app... that reliable petri dish to test in is the most valuable thing."

What reps actually compound

Not specific lessons. Pattern recognition. Speed. The ability to one-shot a sketch in ink without a pencil draft underneath, because you've done thousands of pencil drafts.

Stubbornness on its own is foolish. Stubbornness paired with structured testing is what lets you get more reps in. Some forms of impatience are also good — impatience with bad answers, with surface-level thinking, with settling before you've actually tested something.

What wastes reps

Iterating on the vision when you should be iterating closer to user signal. The difference is subtle but matters: one converges toward something people want, the other converges toward a more elegant version of something people don't.

Each Sail/Muddy iteration started from "how do we make this thesis work" rather than "what are users actually doing." The thesis was so compelling it became hard to question. Reps without enough variable change between attempts, or without putting each attempt in front of users, don't compound.

The petri dish

Bier's framing is the durable one. Your startup should be a petri dish, not a monument. The reproducible process is more valuable than any single idea — because the next idea is going to be different, and the only thing that transfers is your ability to test things fast.

Related

• dogfooding-trap — the kind of testing that doesn't count as a rep
• positioning-vs-vision-gap — why iterating on the vision wastes reps
• no-silver-bullet-llms — the "compelling thesis" failure mode in another domain