Summary

For thirty years, data systems were built for human analysts who supply implicit context, judgment, and skepticism as a matter of course—skills agents completely lack. Fowler argues that the shift to agentic AI demands a fundamental rearchitecting of the data layer across five attributes: trusted, contextual, traceable, governed, and operational. He walks through four engineering disciplines to build each attribute in: data contracts and quality gates (trust), traceability and staged autonomy (governance), a three-part context layer of domain/semantic/capability models (meaning), and a tiered access spectrum from RAG to write-back (action). Observability is treated not as one of four layers but as a cross-cutting constant that must be instrumented from day one. The article concludes that ownership models are as load-bearing as any technology, and that when agents become the primary data consumers, the data architecture simply is the AI architecture.

Key Insight

Every implicit skill a human analyst supplies for free—skepticism, context, institutional memory, judgment about bad data—must now be explicitly engineered into the data architecture itself, because agents act confidently where humans would pause.

Spicy Quotes (click to share)

• 6

  A human hesitates at data that looks wrong; an agent acts on it anyway.

• 7

  Miss one, and the agent won't degrade gracefully the way a person would. It fails confidently.

• 6

  The semantic model doesn't make the agent smarter. It stops it from guessing. For an agent that acts on the answer unchecked, that's what matters.

• 7

  Reversibility predicts safe autonomy better than the size of the transaction.

• 5

  Data quality signals should drive the threshold, not just the model's own confidence. A model can be sure of a stale answer, and the freshness SLA overrides that misplaced certainty.

• 4

  A system you can't explain is one you can't fully trust, defend, or fix.

• 7

  When agents become the primary consumers of your data, your data architecture becomes your AI architecture.

• 6

  Five to ten well described business capabilities will outperform 50 thin API wrappers almost every time.

Tone

authoritative, pragmatic, technical