It’s been a while since I read the bitter lesson by Rich Sutton, with that said, the general “vibe” I picked up from the essay was “building human features and heuristics into a program will eventually be surpassed by general brutish methods”
So this poses the question. Orchestrating workflows using an agent frameworks seems like it goes against the gospel that hacker news users (like myself) preach all so often.
So what do you think about approaching problems with specialize agent flows? Is this something that will shine in many use cases, or become obsolete in the next few years of innovation?
Agentic frameworks are nothing but somewhat impressive glue engineering and hundred percent will be passed over when models with bigger context lengths and better reasoning come about
Rule of thumb: if business folks drop a AI term that becomes the vogue lingo to use, be assured it is a fad that will meet it's doom in days to come
I don't think the bitter lesson is an absolute, but the current investment in agents is what you get when you ignore the bitter lesson