Vibe coding and Serverless are not friends with each other
When non-engineers do vibe coding (GenAI does coding), plain old servers are used, where the worst consequence of inefficient coding is poor performance.
When engineers do vibe coding, the one knowing that Serverless lets us focus on the most important stuff for the business, much bigger problems arise.
Pricing model for Serverless lets us pay for real work and not pay for idle time.
All good except AI hallucinations.
Google Gemini decided that it needs to invoke my AWS Lambda in endless loop😱😱
Thank God I noticed it and shut down everything, otherwise I would have had to sell my kidney.
Or I might be late to notice, AWS has billing delays 🤞🤞
What pitfalls have you encountered with vibe coding?
Solid cautionary tale here. The serverless pricing model amplifies every inefficiency, so AI hallucinations that generate infinite loops turn from minor bugs into financial disasters. Lambda concurrency limits and CloudWatch alarms shoud be mandatory for any GenAI-generated deployments, not optional, because the blast radius is too high otherwise. I've had simlar issues where an LLM missed a base case in recursion and burned through $400 before anyone noticed.