Walden Yan is a Cognition co-founder, listed as Chief Product Officer by AI Engineer’s 2026 conference program. His public work includes neural program-synthesis research, a React wrapper for AI text completion and writing about extending Devin into software operations.[1][4][5][6]
Start with Welcoming TierZero to Cognition →
X Background
Yan’s public GitHub profile describes him as a founder and researcher and records an International Olympiad in Informatics gold medal in 2020. His LinkedIn profile lists Harvard University from 2020 to 2024, but its accessible text does not establish that he completed a degree. His repositories offer a more concrete view of his technical interests, ranging from programming-contest preparation to research on generating programs from descriptions.[3][2][4]
Career
AI Engineer’s 2026 program identifies Yan as Cognition’s co-founder and Chief Product Officer. In his signed July 20, 2026 announcement welcoming TierZero to Cognition, he emphasizes the work required to keep software running after release: detecting incidents and maintaining healthy systems. He says the teams intend to bring TierZero’s automation work into Devin so engineers spend less time handling incidents. The statement establishes his public product direction, while the proposed integration remains an intention in that announcement.[1][6]
Contributions
Teaching a model the structure of programs
Yan’s TokPair-Model repository accompanies the MIT PRIMES project Token Pairing to Improve Neural Program Synthesis Models. Program synthesis means producing code that meets a supplied specification. This project uses English descriptions, example solutions and tests from the AlgoLisp dataset. It studies token pairing as a way to help a sequence model cope with the tree-like structure of programs, alongside attention, search and syntax constraints. The repository supplies preparation, training and evaluation steps. Its reported result belongs to that experimental setup; it is not evidence of unrestricted software-engineering ability.[4]
Putting AI completion into a web form
The EnhanceAI for React repository is a smaller, application-facing contribution. It wraps EnhanceAI so a developer can add AI completion to a text-input component in a React application. The README includes an installation command and a short component example. The source establishes Yan’s repository and the wrapper’s purpose, without proving current service availability or maintenance.[5][3]
Tools, writing & talks
Writing, research & talks
Start here
Welcoming TierZero to Cognitioncognition.comA short signed explanation of Yan’s interest in software operations, incident handling and automation after deployment. This is the easiest introduction to his current product work.
Start with this short signed announcement to understand his current product direction without needing to run research code.
Announcement author[6]
Tools
- TokPair-Model
github.comCode and experimental instructions for generating programs from English descriptions. It introduces the structural difficulties of neural program synthesis and links the accompanying MIT PRIMES paper.
Repository owner and project contributor[4] - EnhanceAI for Reactgithub.com
A React wrapper that adds AI completion to text inputs, with a compact integration example. Useful for understanding the application-facing side of his public code.
Repository owner[5]