Top 10 Kafka mistakes I keep seeing: 1) Treating topics like DB tables: one giant topic, no schema/versioning, then consumers break on deploy. 2) Wrong partition key: hot keys, uneven partitions, 1 partition doing 90% of the work. 3) Assuming ordering across partitions. Kafka only orders within a partition, not across the topic. 4) Auto-commit offsets. Then a crash means you acked work you never finished. 5) No idempotency/dedup. At-least-once + retries = duplicate side effects (double charge, double email). 6) Retrying forever on poison pills. One bad message pins a partition and lag climbs for hours. 7) Not separating retry vs dead-letter topics. Operators end up replaying from raw prod topics. 8) Overfetching payloads: huge messages (MBs) instead of storing blobs in S3 and sending pointers + metadata. 9) Misunderstanding consumer groups: scaling consumers past partition count does nothing except add coordination churn. 10) Weak observability: no per-partition lag, no offset commit metrics, no tracing across producer -> consumer -> DB write