Karpathy is wrong. Write that post, build that slide deck (world.hey.com)

After an interview with Andrej Karpathy sparked a “don’t write, just build” meme, João argues the takeaway is incomplete: building is essential for learning, but explaining — via docs, slide decks, notebooks or talks — is what forces clarity and turns private craft into public impact. He reframes Karpathy’s advice into a compound formula (Luck = Doing × Telling): doing develops expertise, telling multiplies visibility and serendipity. He backs this with workplace examples — a hidden SQL optimization that saved money and improved latency went unnoticed because it wasn’t shared — and describes how modeling and rewarding “build and talk” at Adevinta shifted engineers from quiet shipping to public teaching and conference talks. For the AI/ML community this matters practically and culturally. Sharing artifacts (code, reproducible notebooks, slide decks, postmortems) improves reproducibility, surfaces technical trade-offs, accelerates collaboration, and attracts opportunities that drive careers and projects forward. Leaders can normalize this by example, embedding “tell” into reviews and team rituals so technical wins aren’t invisible. The headline lesson: don’t choose between building and writing — do both. Ship your model, then document its assumptions, failure modes, and evaluation so others can reproduce, critique, and build on it.