He has a Harvard doctorate, a stack of papers on how strangers cooperate, and a neural-interface startup acquired by Facebook on his resume. Ask what he's proudest of and the answer is somebody's first plate of spaghetti and meatballs.
Andrew Mao runs Parsnip, an app his own marketing happily calls "Duolingo for cooking." The pitch fits on a napkin: take the dopamine of leveling up in a video game and aim it at a skill you actually need - feeding yourself. Open the app, and you are not handed a 40-ingredient recipe and wished good luck. You are handed a path. Small wins. Then bigger ones. Then, somehow, dinner.
It works because Mao did not arrive at cooking from cooking. He arrived from the science of how people learn, cooperate, and change their behavior. Parsnip is what happens when someone who spent a decade measuring human decision-making decides the highest-leverage place to apply it is the most domestic room in the house.
The numbers back the instinct. Parsnip reached 4.9 stars on the App Store, crossed 35,000 organic downloads, and got featured by Apple. It went to #1 product of the week on Product Hunt and #1 on Hacker News, sparked five viral threads in cooking subreddits, and landed in Morning Brew. Not bad for an app whose core feature is making you less afraid of an onion.
Mao does not describe Parsnip as a recipe company. He describes it as the first room in a much larger house. The long game is a personalized AI companion for the home kitchen - something that knows your skills, your fridge, and your taste, and quietly answers the oldest question in human history: what's for dinner. Cooking is just the proving ground for a broader idea he keeps returning to - intelligence augmentation, software that helps people learn faster and decide better. It is a strikingly old-fashioned ambition wearing modern clothes: not to replace the human at the stove, but to make that human more capable, one confident step at a time.
Duolingo for cooking.
Cooking is the rare skill almost everyone needs and almost nobody is formally taught. Recipes assume you already know what "fold," "deglaze," and "until fragrant" mean. Cooking shows assume you are watching for fun, not following along. The result is a quiet, widespread anxiety: a fridge full of ingredients and no confidence to combine them. Mao's read on the problem is the read of a learning scientist, not a chef. The bottleneck is not access to recipes. The internet drowns in recipes. The bottleneck is confidence, sequencing, and feedback - the exact things good teaching provides and the exact things a printed recipe cannot.
So Parsnip does what a patient instructor would. It meets you at your level, breaks a daunting dish into a ladder of small, winnable steps, and gives you the satisfaction of a streak. The genius is psychological as much as culinary. A first-time cook who finishes spaghetti and meatballs has not just made dinner. They have proof they can do the next thing. That proof is the product.
The backers tell you how the bet was read. Parsnip's earliest believers include Luis von Ahn and Severin Hacker - the co-founders of Duolingo, the company that turned language drills into a daily habit for hundreds of millions. They did not invest because they love food. They invested because Mao was running the Duolingo playbook on a brand-new skill, and they recognized the moves. Rounding out the early cap table: a James Beard laureate, the kind of culinary authority who lends an app credibility no amount of marketing can buy. Learning science on one side of the table, food expertise on the other, and Mao in the middle, translating between them.
The validation came in public, too. Parsnip topped Product Hunt as education product of the week, hit the top of Hacker News, set off five viral threads in cooking communities on Reddit, and earned a spot among Morning Brew's most popular content of its quarter. For a small team, that is the kind of organic traction that money cannot manufacture - it has to be earned one delighted user at a time.
The bottleneck was never recipes. It was confidence.
He spent years studying whether strangers would cooperate. Now he gets them to cook.
Read his career as a single sentence and it sounds like four different people. Read it as a question - how do humans learn and decide, and how can machines help? - and it is obviously one.
Ask Mao where Parsnip ends and the answer is not "more recipes." In 2024 the team shipped a new generation of Parsnip AI built to "do the heavy lifting" - generating personalized lessons instead of relying on a fixed library, so the app can teach what you want to learn rather than what someone pre-wrote. By 2025 he was writing publicly about a problem most of the industry quietly avoids: building AI tutors that actually work, not demos that dazzle and then fail the moment a real learner gets confused.
That is the throughline of his entire career resurfacing. The PhD on human behavior, the crowdsourcing experiments, the time-series ML at CTRL-labs - all of it was practice for the same question. How do you build a system that genuinely helps a person get better at something hard? Cooking is the wedge because it is universal, daily, and emotionally loaded. But the engine underneath is meant to generalize. Mao talks about an AI companion that knows your skill, your fridge, and your taste, and finally retires the most asked, least answered question in any household: what's for dinner. Solve that with software that teaches rather than merely tells, and the kitchen is only the beginning.
The proudest line on a 1,200-citation resume? Somebody's first home-cooked meal.