The note
In December 2026 I am giving a keynote at ICMarkTech 2026 in Bucharest on practical AI implementation. The thesis is simple: build less demo AI, ship more production AI. Evaluation first, architecture second, governance always on.
In practice that means designing for the parts a demo hides — reliability under real user variability, failure modes when prompts and models change, human oversight in critical paths, and audit-ready documentation aligned with the EU AI Act. It is the same discipline I teach in workshops: a reusable prompt framework, a reference architecture for orchestration and retrieval, a pilot-scoring model to prioritize what is worth building, and governance controls that fit delivery reality rather than slowing it down.
The organizations that win with AI are not the ones with the flashiest prototypes. They are the ones that treat evaluation, architecture, and governance as first-class from day one.