DynamoDB broke the "NoSQL has no transactions" assumption in November 2018. Before that, workloads needing ACID like financial systems or inventory management had to implement complex distributed locks manually at the application layer. The TransactWriteItems API launch enabled atomic writes across multiple tables, breaking that barrier. However, using transactions blindly doubles costs exactly, and not understanding conditional write idempotence means network retries create data duplication. In this day, we dig deep into the 2-phase commit internal workings of DynamoDB transactions, patterns preventing race conditions with conditional writes, and why TTL deletes asynchronously and its implications.