Tooling additions that make agentic data science work possible

​I resisted using LLMs at work until seeing a Data Engineer solve on-call support requests with 2-3 prompts from Slack in early 2026. While still maintaining some skepticism, I’ve since leaned on LLM-based IDEs and apps to parallelize work, dispatch agents for EDA, data quality, and lineage retrieval, build modeling and ETL pipelines, and assist in MLOps work. As LLMs get increasingly more capable, the prompting required to orient them towards data science work feels easier but having harness additions that connect them to data remains essential. Additions like documented (and maintained!) data models (DBT), data lineage tools (Atlan), and plugins for your data warehouse (in this case Snowflake) allow agents to perform longer analyses with less guessing and generate more useful outputs. I’ll also touch on MLOps use cases and how LLMs generally improve the experience of supporting deployed models in production.

​Exact level of detail on specific stories and use cases within my company is still in discussion with my manager, as is whether or not I can consent to recording (I'm checking the box so that I can submit the form). The focus of this talk will likely orient less about the specifics of Engine’s tech stack and tooling, and more about the concepts within that tooling that allow us to deliver on company values like "speed with precision" and "drivers not passengers."

​Key takeaways: Like most data problems, maintained documentation or metadata about your company's data is still the foundational pillar for making your employees and your agents lives easier.