Dmytro Yaroshenko

An agent produces a plausible change and says it is done. What evidence would make you trust it? This talk follows the move from personal use of engineering agents to shared team workflows and continuous automation. We will look at the problems that remain difficult: useful feedback from the environment, validation across long tasks, shared ownership, and the cost of getting work accepted. Attendees will leave with a way to choose their next bounded workflow, define its evidence of success, and decide where human approval still belongs.

dbt recently announced a multi-engine data mesh strategy which is appealing to many. In this session, we will discuss embracing a pragmatic approach to multi-engine landscapes. For edge cases, Snowflake can govern additional engines via Snowpark Container Services (SPCS). We will demo a dbt project executing models across Snowflake warehouses and Dynamic Tables alongside DuckDB in SPCS as an example of an external engine. Results will be stored in Parquet/Iceberg, governed by Snowflake Materials.