AWS DynamoDB recently experienced a significant outage that disrupted many services online. Reports included bizarre scenarios like people stuck in sleeping pods and smart locks failing due to the outage. AWS attributed the issue to a race condition in its DNS management system, leading to increased API error rates in the us-east-1 region. The author suggests that such issues may seem obvious in hindsight, but many potential problems are overlooked until they escalate. The article advocates for using System-Theoretic Process Analysis (STPA) to identify and mitigate risks in software systems, even those that aren't safety-critical. The author describes performing a light STPA analysis on the AWS summary report, finding that it not only uncovered the problems leading to the outage but also revealed other issues that could cause future outages. Given the estimated financial impact of the outage—between $40 million and $600 million—regularly applying STPA during development could be a cost-effective strategy. The STPA method involves defining the analysis's purpose, modeling the control structure, identifying unsafe control actions, and developing loss scenarios. The author emphasizes that unsafe control actions can stem from either harmful inputs from controllers or failures to provide necessary control inputs. This approach allows for a more thorough examination of potential failure modes compared to traditional root cause analysis, which the author criticizes for its limitations.