It’s likely a mix of availability bias and pattern-seeking/apophenia: a few recent outages stand out and the rise of LLMs in recent years is also ever present. This increases the chance to assume a link between both even without evidence that outage frequency actually changed.

Also, I don’t think the switch is as binary: these industry shifts tend to be more gradual where more experienced programmers adopt LLM tools to suit their workflow and newer developers still learn fundamentals because real-world engineering of complex systems quickly exposes a lack of understanding.

The way I think about it is that LLMs are just a tool, and if you trust the tool too much it can backfire. It reminds me of this video [1] that Louis Rossman posted regarding a police officer essentially trusting his AI tool (Flock cameras) too much and falsely accusing a woman of a crime, claiming "you can't take a breath of fresh air without us knowing about it".

[1] https://youtu.be/AoEQg1M92_E?si=A-XNXP_smH2I3hWj

I think you're looking for a pattern. Cloud's been widely criticised for price and reliability since the term was coined, including via classic jokes https://xkcd.com/908/ and https://archive.org/details/i-just-want-to-serve-5-terabytes