Building data agents

For years, "AI for data" just meant a generic text-to-SQL chat box. Lately, though, we're finally moving toward true Data Agents which are systems that use closed execution loops to autonomously plan, run queries, check outputs for anomalies, and self-correct. I would LOVE to understand what are the tradeoffs of creating your custom data agent vs using one already built out of the box like Snowflake Cortex analyst, Databricks Genie, or PowerBI copilot. ​The platform-native stuff is actually getting pretty interesting. If you look at what Databricks just rolled out with Genie, they’ve shifted it from a basic Q&A interface into a full autonomous agent space by embedding it directly into their governance and Genie Ontology framework. Because it sits natively on the metadata, it acts less like a glorified autocomplete and more like a junior data analyst who inherently understands your table relationships and guardrails. It also has a built in harness. Is it worth designing your own AI Data agents using langgraph still or are managed agents the way to go? Thanks!!