I don't know... Maybe I spent too much time studying how to tame AWS using IaC and gitops reproducible deployments, but AWS lambda seemed to me the most impressively simple and inexpensive product. Once I did an complete project, from end to end, designing the architecture and flow of multiple lambdas communicating with each other through SQS queues to search, extract and load info from geotiff files from S3 into a PostgreSQL database, and it was really straightforward.
If you leverage docker images for deployment and separate the interface for treating lambda requests from the core logic, it doesn't have much space for surprises.
If the author went with the cliché that lambda scalability can harm your budget, it wouldn't be original, but at least it would have been plausible, but complex? I don't know, maybe someone could present the case with more deails for why it's so complex.