You're being told two stories about agentic coding on Databricks, and they both miss the point. Lakeflow Designer says anyone can build pipelines now with just some simple prompts. Genie Code promises to automate building anything and everything. The reality is messier and more interesting: there's a four-tier maturity curve, most teams are stuck at tier two, and the painful gap between "we use Genie Code" and "we've actually built engineering opinion into Genie Code" is where you're going wrong. So where are you on the curve; and where do you think you are?

This session walks the four tiers as a maturity curve: from prompt-by-prompt low-code (Lakeflow Designer), through unharnessed Genie Code, through Genie Code shaped by custom skills and engineering opinion, to full bespoke agentic development frameworks. We'll run through each tier, see what it's good for and where it falls over. We'll spend a lot of the session looking at how you build out an opinionated set of skills, instructions and best practices inside Genie Code, because that's the tier most teams think they're at and almost nobody actually is. We cover what a real skill abstraction looks like, the layered architecture beneath it (Databricks mechanics, opinionated patterns, consulting craft), and what changes when you operationalise this across an engineering team rather than running it solo.

If you're a data engineering leads, architect, or senior practitioners who suspect the agentic tooling story is more complicated than the vendors are letting on, this session is for you. You'll leave with an understanding of what good looks like and concrete patterns for moving up the curve without setting your engineering practice on fire.