
AI Agents in Production: The Eval and Observability Playbook
Models fail loudly. Agents fail quietly — wrong answers, drifting tools, stale context. The eval and observability stack that makes them production-grade.
Field notes & perspectives
Honest writing on data strategy, platform architecture, analytics and AI — drawn from programmes we've led with banks, healthcare, fintech and consumer brands.

Hard lessons, specific thresholds, and operational patterns that actually work when running Iceberg at scale in 2026 — not just on a slide.

Gartner expects 60% of AI projects to be abandoned through 2026 without AI-ready data. Why lakehouse foundations are the leverage point for leadership teams.

AI agents fail because data foundations are broken. The governance-first blueprint we use across 30+ engagements to make AI work in production.

Metric drift and broken AI agents are symptoms of the same problem. Why data contracts and a semantic layer are the foundation that fixes it at the root.