Jay had opinions about READMEs.
This was not news to anyone who had worked with him. Jay had built html2text, a Go library that converted HTML to plaintext, and its README was a minor masterpiece of technical documentation—clear, structured, example-rich, and ruthlessly honest about the library's limitations. He believed that a README was a contract between the author and every future developer who would encounter the project at 2 AM while debugging something unrelated and desperate for a tool that just worked.
So when he decided to open-source a Go client library for interacting with CXDB, the code was the easy part. The agents wrote the code. The scenarios validated it. The satisfaction metric climbed past 0.96 in three days. But the README was Jay's.
He wrote it by hand. Every word. No agent assistance. It was the one piece of writing in the entire factory that a human insisted on producing, and nobody argued with him about it, because they'd seen what Jay could do with a README.
The library was called cxdb-go and it did exactly what the name suggested: it let Go programs interact with CXDB. Create contexts. Append turns. Branch from any point in the DAG. Query the immutable history. Stream changes in real time. Everything the CXDB binary protocol supported, wrapped in idiomatic Go with proper error handling and context cancellation and all the things that Go developers expected because Jay was a Go developer and he expected them too.
The README opened with a single sentence: cxdb-go is a Go client library for CXDB, the AI context store. No marketing. No hype. Just what it was.
Then examples. Six of them. Each one a complete, runnable program. Each one demonstrating a single concept. Create a context. Append a turn. Branch a conversation. Query the history. Stream live updates. Handle errors. Jay had tested every example by copying it into an empty main.go file and running it. If it didn't compile and produce the documented output, it didn't ship.
He published the library on a Thursday. By Friday, two internal teams at StrongDM were using it. By the following Monday, three external teams had opened issues—good issues, thoughtful issues, the kind of issues that meant people were actually reading the code and using it in production.
Within a month, twelve teams were using cxdb-go. The issue tracker was active. Pull requests arrived. Jay reviewed every one personally, because the README had set a standard and he intended to maintain it.
Navan asked him once why he insisted on writing the README by hand.
"Because the README is the first thing someone reads," Jay said. "And first impressions should be human."
Justin, overhearing, said nothing. But he smiled. There were some things the factory couldn't automate. Some things it shouldn't.
Jay testing every README example by copying it into an empty main.go. That's the standard. That's the bar. Every open-source maintainer should read this story and feel appropriately convicted.