Are domain-specific agents quietly beating general coding agents in their own niche?

After a few weeks of testing, it became clear that the moat for agents isn't the model, it's the context. General coding agents (Cursor, Claude Code, etc.) are great at code but blind to data. I've been playing with Databricks' Genie Code and the difference isn't intelligence — it's that it already knows the schemas, lineage, and permissions before it writes a line. It fails less because it guesses less. However, the potential downside, it works inside the platform (i.e. the opposite of portable). Curious if anyone's seeing the same pattern in other domains — are vertical agents with native context winning against horizontal agents with better raw capability?