@stefano coming from a different approach, many of the llm models know bsd internals and approaches just fine. i launched an agent (we fully control the gpu cluster at the dc) at a jail cluster staging candidate and it found the vnet misconfiguration in less than 30s.
i'm also seeing more and more companies say "we'll use ai" but the funny part is the clients i'm starting to lose to ai/llm come back with problems they can't solve since they don't have technical staff. as a result i reset my rate schedule with them at a higher cost. two have still asked about llm use, so i've shifted to building them private local-only (no calls) ai clusters for automating a lot of their tasks (think lead-gen, marketing analysis, financial tedious tasks)
i think it can be used in a constructive way, but companies see "ai/llm" as this magic wand they can wave which most shoot themselves in the foot.
i realize it's not a pivot for everyone, stefano but so far it's worked for me.