Queues for Kafka has been a long-requested primitive — the ability to consume messages in a queue-style (at-most-once, competing consumers) from a Kafka topic without building consumer-group gymnastics on top. Shipping it at GA in 8.2, alongside Flink SQL queue support, rounds out a pattern that a lot of stream-processing teams have been hacking together themselves. The managed MCP server is the more forward-looking piece: it lets AI agents interact with Kafka clusters through the Model Context Protocol rather than custom tooling. Whether that's immediately useful to your team probably depends on how far along your agentic infrastructure is, but the integration layer existing at the platform level is worth tracking.