Salma Bakouk

Every data estate drifts. Schemas change, definitions fork, freshness slips, and the gap between what the data says and what its consumers believe widens over time. For the last decade we have managed that gap with people. Someone notices the number looks wrong, asks in Slack, and the correction happens before the decision does. Agents remove that person. They read the table, decide, and act, on a loop, with no pause where a human could have caught it. The drift is the same drift. What changed is that there is no longer anything absorbing it. This panel looks at what that means in practice. Who is accountable when an agent acts on a broken table. Whether contracts defined on the asset are enough, or whether each consumer needs its own tolerance. And what a reliability check looks like when it has to run before the action rather than after it.

AI is only as strong as the data beneath it. Yet most enterprises still rely on fragmented tools and reactive processes that undermine trust. The result: innovation that looks impressive in demos but collapses under real-world pressure. In this keynote, Salma Bakouk, CEO of Sifflet, argues that metadata, not models, is the true foundation for the AI era. By building a metadata control plane enriched with agentic observability, enterprises can move from reactive patchwork to proactive intelligence. In this keynote, she offers a provocative vision of where the market is heading, what traditional approaches are getting wrong, and why the winners of the AI economy will be those who treat trust not as insurance, but as infrastructure.