The discussion centers on the evolution of data platforms from passive 'systems of intelligence', which historically saw only 10-20% of insights operationalized, to active 'systems of action'. Generative AI and agentic systems now bridge the gap between insight and execution, allowing for the direct automation of business tasks based on real-time data.
A core argument is that traditional data quality metrics (cleanliness, lineage) only account for 50% of an AI agent's accuracy. The remaining 50% comes from rich business context—the 'invisible work' and intuition previously held only by human experts. Google's Knowledge Catalogue aims to codify this context by inferring schema and meaning across both structured and unstructured data.
The conversation highlights a trend away from monolithic, persona-based agents towards 'swarms of agents' that collaborate to fulfill a user's high-level intent. This enables a new development philosophy of 'intent-driven engineering,' where practitioners focus on objectives and outcomes, while agents, equipped with tools and skills via the Data Agent Kit, handle the underlying tasks.
Google Cloud is addressing the multi-cloud reality by enabling data access and analysis without requiring data movement. Leveraging open standards like Apache Iceberg and services like Cross-Cloud Interconnect, customers can create a unified data lakehouse that queries data in-place across AWS, Azure, Databricks, and Snowflake.
As AI agents generate 10-20x more API calls than humans, managing cost and performance becomes critical. Google emphasizes its unique ability to innovate and optimize across the entire stack—from custom TPUs that separate training and inference workloads to massive performance-per-dollar improvements in BigQuery and Spark.
Keep pulling the thread on Yasmeen Ahmad.