350 CTOs Share Playbook for AI-Native Engineering at Snowflake Summit

Snowflake's inaugural CTO Circle, held during Snowflake Summit 2026 in San Francisco, convened more than 350 CTOs from financial services, telecommunications, retail and technology to address a fundamental challenge facing engineering leaders: how to build AI-native organizations when no established playbook exists. Vivek Raghunathan, SVP of Engineering at Snowflake, framed the transformation not as a tooling problem but as a management philosophy shift. Rather than assuming leadership knew what engineers needed, Snowflake applied product management principles to internal developer experience—interviewing engineers to understand friction points, establishing baseline metrics, running experiments to gauge impact, and combining executive sponsorship with bottom-up adoption.

The results were concrete. In 18 months, Snowflake increased its internal developer Net Promoter Score by more than 30 points, yielding a 4:1 ratio of satisfied to dissatisfied developers. More importantly, that improvement translated into faster software delivery and quicker adaptation to evolving AI capabilities. Raghunathan identified three stages of AI adoption: adoption (when developers learn tools), mastery (when engineers discover repeatable workflows that consistently produce better outcomes), and optimization (when those workflows become organizational knowledge). At Snowflake, early AI adopters experimented with prompting techniques, planning methods and debugging approaches, and the organization then documented successful patterns and made them available across engineering. Engineers using proven workflows consistently outperformed those at early adoption levels—a finding that shifted the competitive focus from tool deployment to workflow institutionalization.

Operational discipline emerged as a second critical theme. Jeremy Burton, General Manager of the Observability Business Unit at Snowflake, argued that AI has entered a production phase where organizations must demonstrate measurable business value, not isolated technical successes. As AI agents generate more telemetry and interact with more systems, observability shifts from monitoring infrastructure to providing context for reliable AI operation. A common enterprise assumption—that better models simply need better data—misses the point: AI is only as effective as the context it can access, including semantics, relationships captured through ontologies and knowledge graphs, and business context. Aditya Gaur from Netflix demonstrated this principle through automated root cause analysis. Rather than a pure AI initiative, its success depended on years of prior work connecting fragmented telemetry, modeling operational relationships, and creating a shared context layer. By the time AI was introduced, the foundation already existed for agents to reason over structured operational knowledge and generate meaningful incident hypotheses.

The discussion also surfaced tensions between speed and governance. Caitlin Colgrove, CTO at Hex, described full commitment to AI—"burning the boats"—as necessary when organizations reach scale. Hex initially created a dedicated AI product team, but this created bottlenecks; the company ultimately disbanded the centralized team and distributed AI responsibility across every product team, enabling faster delivery and broader capability embedding. Conversely, Chris Kozlowski from Barclays emphasized that in regulated environments, speed must be matched with governance and trust—engineering teams cannot choose between innovation and control. Corey Burke and Arun Rajamanickam noted that as AI agents become capable of implementing significant portions of features independently, engineers increasingly spend time defining intent, orchestrating agents, and validating outcomes rather than writing code. Success requires engineering platforms that enable rapid experimentation without compromising reliability.