In Season 4, Episode 38, Karl and Jon welcome AWS Community Builder Luis Valdivia. They discuss the new ECS managed instances for containerized applications on AWS, the cost-effectiveness of serverless architecture at scale, and the latest Anthropic Claude Sonnet 4.5 model now available in Amazon Bedrock. They also cover the general availability of the AWS Knowledge MCP server and a recent social engineering attack on a software platform that exploited AWS domain registration. The episode wraps up with a lighthearted moment as the hosts spend ten minutes trying to determine whether Jon’s background had actually changed or not.

06:05 - ECS managed instances for containerized applications 

AWS introduced a new way to run containers called ECS managed instances. This option sits between unmanaged EC2 instances and Fargate, offering more control than Fargate but less management overhead than unmanaged EC2. The pricing model is based on instance type, with a management fee added. This new option provides more flexibility but also adds complexity to the decision-making process for container deployment.

16:17 - Serverless cost-effectiveness at scale 

An article by AWS hero Evandro Pires argues that serverless is not inherently expensive at scale, but rather becomes costly when implemented incorrectly. The discussion highlights that serverless encompasses more than just Lambda functions and that proper architecture is crucial for cost-effective serverless deployments. The speakers agree that bad architecture, rather than the serverless approach itself, is often the root cause of high costs.

23:44 - Anthropic Claude Sonnet 4.5 model in Amazon Bedrock 

AWS announced the availability of Anthropic's latest AI model, Claude Sonnet 4.5, in Amazon Bedrock. This model is described as Anthropic's most intelligent, particularly for coding and complex agents. The speakers discuss the benefits of having this model integrated into AWS's ecosystem, including enhanced security and data privacy. They also note the incremental improvements over previous versions and its capabilities compared to other models.

30:11 - AWS Knowledge MCP server 

AWS released the Knowledge MCP server, which allows large language models (LLMs) to access AWS documentation and knowledge bases. This tool aims to reduce hallucinations and provide more accurate information when using AI for AWS-related tasks. The speakers highlight its potential to improve the reliability of AI-generated code and documentation for AWS services.

36:14 - Social engineering attack on AWS domain registration 

A software company called Kodex experienced an outage due to a social engineering attack that targeted their domain registration through AWS. The speakers discuss that while the attack occurred through AWS's systems, it's not entirely fair to blame AWS as it was a human vulnerability rather than a technical exploit. They suggest that AWS and other providers may need to tighten verification policies for domain management, especially for high-profile targets.