What enterprise AI agents need to work reliably in production

A panel from the Imagination in Action Silicon Valley AI Summit explores the technical and organizational foundations needed to deploy AI agents inside companies. Speakers include Deloitte's chief data officer, founders of data-infrastructure startups WEKA and Alation, and a Stanford management science professor.

Key topics include data catalogs, vector databases, knowledge graphs, and access permissions—the infrastructure layer that determines whether agents can act reliably. The group discusses why evaluation and monitoring matter, how data drift breaks automation, and the problem of high automation rates creating new human oversight challenges.

A central theme: domain experts and human judgment remain essential when automated workflows encounter exceptions or edge cases that the training data didn't prepare them for.