Anthropic outlines a four-layer analytics architecture: data foundations with governed models and metadata, the knowledge layer with semantic definitions and business context, skills encoding repeatable workflows, and validation systems that ensure correctness and consistency.

Claude now handles roughly 95% of internal analytics requests, enabling employees to query business data independently instead of relying on data teams.

Reaction from the data community is mixed, with some praising openness while others push for deterministic, idempotent analytics results.

The approach tackles self-service analytics challenges such as overlapping datasets and conflicting metric definitions, and aims to support long-tail business questions without flooding dashboards.

An appendix includes a redacted template of the skill file that guides analytics agents.

Ultimately, success comes from data governance, semantic definitions, and disciplined operations, not just advances in models.

Key principles for AI-driven analytics include maintaining a single source of truth for metrics, ensuring easy access to the right data, and continuously detecting stale definitions.

Semantic metrics, data lineage, query patterns and business context are viewed as the true sources of truth for analytics agents, supported by structured definitions and human-owned documentation.