I had the opposite experience with Lambdas/Serverless. It made easy things very slightly easier, and hard things impossible.
AWS-hosted Redis? Sure! AWS-hosted RabbitMQ? Of course! AWS SQS? No thanks - it wouldn't give me anything useful that RabbitMQ can't do, however it will lock me into forever having to pay AWS, deal with their terrible console (or stacks of YAML files) to control the thing and expose me to potential edge-cases I will have no way to debug or replicate locally.
Allow other AWS services to send it events (SNS, lambda, etc etc)? Reliably handle millions of messages per second without managing anything yourself? Transparent KMS encryption? Fine-grained role based permissions via IAM without any hard-coded credentials?
Come on.
The other axis is "do they dogfood?" meaning "if I go to amazon.com is the service used?"
The services Amazon built for themselves - s3, rds, ec2, dynamo, ... are generally pretty good. Redshift and glue? Not so much.
For me, the convenient "Lego blocks" that I use a lot include CloudFront (web server/CDN), S3 (file/web server), ACM (certificate manager), DynamoDB (NoSQL database), Firehose (message batching and storage), SQS/SNS (queuing and publish/subscribe), Step Functions (low-code business logic), Lambda (event-driven custom code execution) and API Gateway (REST API / WebSocket service). It is very rare to need to run any self-managed servers any more with all these services available.