Agentic AI · Aug 2026
What is an AI agent — and what it isn't
A working definition — agents own tasks within guardrails — and the test that separates agents from assistants and automation.
Read the noteNotes
Field notes on building AI in production.
Short, practical notes from talks, workshops, and delivery work — on conversational AI testing, LLMs in production, agentic workflows, data spaces, and governance by design.
- Conversational AI Testing
- LLMs in Production
- Agentic AI
- Data Spaces
All Notes
Latest field notes.
LLM Evaluation · Aug 2026
Judge models: the quality gate for AI systems
What LLM-as-a-judge is, where it belongs in an evaluation stack, and why an uncalibrated judge is just a second opinion you can't trust.
Read the noteLLM Evaluation · Aug 2026
LLM regression testing: catching silent drift
Why LLM behavior drifts silently after model and prompt changes, and how golden-suite baselines turn releases from bets into decisions.
Read the noteConversational AI · Aug 2026
Testing rigor is becoming the real differentiator in conversational AI
Why repeatable scenario suites, baseline comparisons, and drift checks separate systems that ship from systems that break after a model or prompt update.
Read the noteProduction AI · Aug 2026
Production AI, not demo AI — the thesis behind my ICMarkTech 2026 keynote
Evaluation first, architecture second, governance always on: the framing behind the ICMarkTech 2026 keynote on practical AI implementation.
Read the noteData Spaces · Aug 2026
Data spaces and AI: trust and sovereignty by design
How data-space-aligned architecture and governance controls make multi-organization AI collaboration compliant and stable — notes from the Ocean Enterprise work.
Read the note