I spent yesterday with langchain tutorials. It works. I hated the workflow because I hate python project and build issues and using jupyter w/o tab completion. I suspect I can find a jupyter IDE with vim bidings. Or I'll use Rust and llm-chain.
But it is a waste of time - the messyness and fragility of gaming prompts to get something needed in an intermediate step shows me what debugging will be all about and how fragile the thing is and the moronic assumptions everyone are making - these working prompts found / gamed during dev are not deterministic and will break and fuck us all if the app is doing something that matters.
So the reason I'm going to seem like a jerk for not playing is we have massive unused resources qued up to spend years failing at this stuff and won't be noticably worse off for years. Enterprise IT is sooooo wasteful and all those resources that had been working on failing projects pivoting to this nonesense won't matter ... wasting IT trillions in a new way is a zero sum game.
... except that we may not wake up from this this time and we'll start watering our crops with sports drink.