Individual Kafka consumers and producers tend to have simple behavior, which is good, and initially plugging them together yields simple, predictable systems, but people tend to keep going until they experience pain, at which point they have a system that is right at the limit of what they can understand. Then further feature work pushes them over the limit into darkness.

I think Kafka needs the equivalent of OpenAPI and Redoc, a simple spec and document generator, but for groups of consumers and producers rather than single applications. This would increase the tractability of complex systems, but it would also let you see the system getting more complex over time, even when you haven't reached the pain point yet.