10 Lessons from 10 years at GitHub
While I contributed to GitHub’s Rails monolith, most of my work focused on building code intelligence services external to it.
This included contributing to Tree-sitter grammars and runtime, helping build a research-oriented program analysis library in Haskell called Semantic, and shipping GitHub’s first code navigation service powering jump-to-definition and find-all-references features.
With the rise of generative AI, I contributed to Copilot Chat for the web and authored the prompt-building library behind its dynamic context logic.
People often describe “moats” in terms of features, data, or network effects. But none of that matters if your foundation is broken. A product with 1,000 features has little value if 900 are buggy or slow.
Before GitHub’s 2018 acquisition by Microsoft, the Dear GitHub letter captured widespread frustration with the platform. Under Nat Friedman’s leadership, a “paper cuts” initiative helped rebuild trust with the open source community by fixing small but painful issues.
