Forty-seven technical writers at Snowflake's Redwood City office spent their final weeks on the job doing something grim: teaching an AI system how to do their work. Then they were walked out with two weeks of severance.
According to accounts from affected employees, Snowflake ran an eight-month process that included screen recordings of documentation sessions and six weeks of formal "knowledge transfer" where senior writers trained the AI pipeline on their research methods, writing style, and editorial judgment. The company deployed what it described internally as achieving "300% efficiency gains." Three contractors in Poland now handle work that previously required the full 47-person team in California.
One 12-year veteran of the documentation team described the experience bluntly: they spent three months teaching the AI how they think, how they write, and how they research. They built their own replacement.
Snowflake confirmed "targeted workforce reductions" but hasn't publicly disclosed the full scope. Internal December meeting notes reportedly described the process as "extraction phase complete, human redundancy achieved." The affected writers' manager was promoted to "Head of AI-Driven Content Strategy" the same week leadership announced plans for 40% cost reduction in non-engineering roles by Q3.
The Uncomfortable Template
This isn't the first time a company has asked workers to train their replacements, but the mechanics here are unusually explicit. Most AI-driven layoffs happen after a tool gets good enough on its own. Snowflake's approach was more direct: use the humans as training data, then remove the humans.
The documentation quality has reportedly stayed stable since the transition, which is the detail that should unsettle anyone in a knowledge work role. The standard objection to AI replacing writers is that quality drops. When quality holds because the AI was literally trained by domain experts with a decade of institutional knowledge, that objection disappears.
For anyone working in technical writing, content strategy, or documentation, this is a case study worth studying closely. The pattern of "help us build this system" followed by "we no longer need you" will repeat across industries. The question is whether companies will be this transparent about it, or whether most will do the same thing without the explicit knowledge transfer sessions.