Orchestration will likely need to be part of any data stack as it grows in maturity, but IMO you don't need to worry about it for a long time unless there's a very good use case for it - e.g. if ML is core to your business, if you absolutely need custom data pipelines, etc. dbt alone should replace core orchestration use cases for a long time.

But agreed that this is only half the picture. There are components to this I've purposely left out for now: orchestration, but also the metrics layer (too early to see how this'll develop, and dbt may solve this as well), monitoring/observability/CI for BI, the whole world of ML deployment/monitoring/retraining tools. But to me these are auxiliary value adds to the core immutable components that you need early on.