Snowflake has eliminated its entire technical writing and documentation department — approximately 70 employees — and replaced them with an AI system called Project SnowWork. The company says the new system generates API documentation and user guides “directly from source code in minutes,” work that previously took human teams weeks.
This isn’t a reduction. It’s a complete replacement. And it follows a disturbing pattern: the company reportedly screen-recorded its technical writers for eight months before laying them off, using the footage to train the AI that would take their jobs.
How It Happened
The layoffs follow Snowflake’s $200 million partnership with OpenAI, announced in February, which integrated GPT-5.2 into Snowflake’s AI Data Cloud. Project SnowWork is the first major application of that partnership.
According to reports, senior writers spent their final six weeks actively transferring knowledge to the AI system. The eight-month recording period captured documentation workflows, turning years of institutional knowledge into training data. Management now claims “300% efficiency gains” from the AI-powered pipeline.
Project SnowWork uses role-based AI personas that understand business workflows, terminology, and KPIs. It comes pre-configured for finance, sales, marketing, and operations, with the ability to generate quarterly business reviews, pitch decks, and customer communications. The platform inherits existing access controls and audit logging — every action is traceable.
The Industry Reaction
Developer experience expert Dachary Carey called the decision “absolutely bonkers,” noting that technical writing teams are how companies ensure people can discover and learn their products.
The concerns are practical, not just philosophical:
Documentation debt. AI-generated guides can hallucinate. In API documentation, a hallucinated endpoint or incorrect parameter can waste developer hours. In security documentation, missing edge cases can create vulnerabilities.
Institutional knowledge. Technical writers don’t just document what code does — they translate engineering decisions into language users understand. That translation requires context that isn’t in the source code.
Quality feedback loop. When documentation is bad, human writers hear about it through support tickets, user feedback, and internal complaints. AI systems don’t have the same feedback mechanisms, and Snowflake’s 12,600 customers will discover gaps only when they hit them.
Snowflake’s Bet
Snowflake is betting that AI can match human documentation quality while dramatically reducing costs. The company’s research, ironically published just days before the cuts, found that 77% of organizations report increased hiring from AI adoption.
That research didn’t ask whether the new hires were replacing the people AI displaced.
The Broader Pattern
Snowflake isn’t alone. Amazon, Canva, Atlassian, and others have made similar moves in recent months. What distinguishes Snowflake is the completeness of the replacement — not a reduction, not a restructuring, but a wholesale elimination of an entire function.
It’s also one of the most transparent examples of employees training their own replacements. The eight-month recording period and six-week knowledge transfer weren’t hidden. They were policy.
Who Wins, Who Loses
Snowflake wins if Project SnowWork maintains documentation quality while cutting costs. Shareholders certainly liked it — the company framed the move as part of becoming “operationally lean while accelerating AI offerings.”
The 70 technical writers obviously lose, along with the profession’s broader confidence that specialized knowledge protects jobs. If senior writers at a major tech company can be replaced after an eight-month surveillance period, the model is replicable.
Snowflake’s customers face the real test. Enterprise software documentation isn’t blog content — errors have downstream costs. If SnowWork halluccinates a security configuration or misses a critical deprecation notice, the efficiency gains will evaporate in support tickets and production incidents.
For now, this is the most concrete example yet of what “AI replacing jobs” actually looks like in practice: surveillance, knowledge extraction, and layoffs, executed systematically over months.