Every AI project begins with a demonstration that impresses. Most stall somewhere between demo and production. The industry has been asking which model is best? – when the answer lives elsewhere entirely.

Frank Weigel, Chief Product Officer at Matillion, makes the case that 2026 is the year of context: the year it becomes undeniable that AI systems are only as intelligent as what they know about your specific business. The failure mode that kills enterprise AI is not hallucination. It is a confident, plausible answer built on the wrong data — and it’s undetectable without the right infrastructure around the model.
Attendees will hear why human-curated approaches like data catalogs have always failed at scale, what a properly architected context system looks like, and leave with a three-question diagnostic for evaluating any AI initiative they’re running today.

Takeaways

• Enterprise AI failures stem from missing company-specific context for data selection, not weak models — and the most dangerous outputs are structurally correct but semantically wrong, which conventional testing can’t detect
• Human-curated context systems like data catalogs are structurally flawed; context must be AI-managed and continuously updated to work reliably
• Trustworthy AI depends on two things: sufficient context for correct data selection, and lineage-based transparency that domain experts (not just engineers) can verify
• A simple three-question diagnostic can assess the context maturity of any AI data initiative.