The Optimization Paradox

Most data teams are not short on insight anymore. The dashboards are full. Cost reports flag cost overruns. Observability platforms catch infrastructure misallocations. New AI assistants will even draft query rewrites for you.

And yet, quarter after quarter, platform spend keeps climbing. Inefficient workloads keep running. The optimization backlog gets longer, not shorter.

The gap is not knowledge. It is execution. Engineering teams already know more than they have time to act on. Every new recommendation is one more line on a list that competes with feature work, SLA fires, and the next migration.

This session takes that problem seriously. We unpack why the most common approaches stop short, what it would take to actually close the gap, and where AI is finally starting to do meaningful work on the other side of the recommendation.

‍

• Why four common approaches (observability platforms, compute rightsizing, AI copilots, and in-house tools) keep producing the same result: more recommendations, not more fixes.
• What changes when an AI system has real context across queries, pipelines, compute, and storage, and how that context is the difference between a suggestion and a safe, validated change in production.
• What autonomous data platform operations look like in practice, with a preview of the work happening today on Databricks, Snowflake, and BigQuery and the early results teams are seeing.
  ‍

Data platform leaders, data engineers, and FinOps practitioners responsible for the cost, performance, or reliability of Databricks, Snowflake, or BigQuery environments. Especially relevant for anyone who is tired of being shown the problem without any practical way to actually fix it.
‍