Why Snowflake kept its FDE team inside product: its Head of Agentic Transformation on "frontier engineering"

Snowflake started putting its own engineers inside customer environments roughly 18 months ago, after finding that strong models and a strong product still weren't getting customers to outcomes. In episode 4 of Contour's The Last Mile podcast (Oct 13), Luv Kothari, Snowflake's Head of Agentic Transformation, explains why the company kept this "frontier engineering" team inside product and engineering, while OpenAI, Anthropic and Microsoft set up separate deployment arms.

Kothari draws a hard line between this and professional services: solutions architects advise or build a light proof of concept, whereas Snowflake's FDEs work hands-on in the customer's environment with the customer's data, on outcome-based engagements of three to six months before handing off. Every project has to pass three tests: real customer impact, repeatability (80–90% of the same core counts as a win), and a long-term right to win for Snowflake. The team mixes deployment strategists, product managers focused on what can be repeated, and AI-native engineers, many of them early Cortex AI engineers.

He also shares a concrete technical lesson: a frontier model wired straight to Snowflake tables scored about 25% on hard data tasks, but around 85% with a proper context layer, while using half the tokens. Snowflake first applied the approach to itself with Raven, an internal go-to-market assistant that grew from about 10 tables to hundreds and serves a 5,000–6,000-person GTM org.

Why it matters for FDEs: it's a clear inside look at how an FDE team can sit within a product org, where field work feeds the core product directly and scoping for repeatability is part of the job. Snowflake is currently hiring for several FDE roles, listed below.