Databricks’ token consumption kept rising while costs stayed flat, Ali Ghodsi says in interview excerpts shared by Sarah Wang. He describes introducing budgets and warnings for individuals and groups, then using analytics to predict where spending was heading. On adoption, a16z’s summary of the conversation says Ghodsi believes most companies already have sufficiently smart models; what those models lack is the organizational context employees absorb over years. On risk, a16z says Ghodsi doubts a runaway self-improvement loop while frontier training requires ever more resources. Cybersecurity is his bigger concern: he believes most organizations are unprepared for the coming changes in AI agents’ capabilities.