We faced this issue during one of our projects and spent at least two retrospectives on how we can fix it in-house, so it is a huge relief that we can use it, and even tailor it to our needs later.
As a developer, I highly recommend the Serverless Icebreaker utility. It's a very practical tool that addresses the challenge of cold start duration in AWS lambdas. It can be used effectively to reduce latency and enhance performance useful to identify early the problems with the lambda size analysis, lib usage identification, and threshold errors, it alse provides actionable insights for optimization. Integration into projects is straightforward, and the tool offers valuable metrics. It's a valuable asset for optimizing AWS Lambdas and delivering improved user experiences.
I've tried this on my SST project - luckily all my lambdas were ideal size, but for the future I've added it to the pipeline - wondering if it will catch something :)