Anthropic automates 95% of analytics queries with Claude and shares its agentic stack

Anthropic reports automating 95% of its business analytics queries via Claude with roughly 95% aggregate accuracy, freeing data scientists for strategic work. The company identifies three core failure modes for analytics agents: concept-to-entity ambiguity, data staleness, and retrieval failure. To combat these, Anthropic built an agentic stack emphasizing canonical governed datasets, enforced standards through tooling and CI, and reusable "skills" that map user questions to precise data entities. The post argues that analytics accuracy is fundamentally a context and verification problem rather than a code-generation issue, offering a template for data teams adopting LLM-driven self-service insights.