my article on the topic: https://t.co/bjWtFe4Jqi our bootcamp:

I'm telling you from experience: almost all interesting cases are in the gray area between agents and workflows...

this oversimplification trend is the thing that sells, or maybe it's a genuine attempt at creating discourse. you pick based on how cynical you want to be. as the guy way smarter than me once said: "Life isn't black and white. It's a million gray areas, don't you find?"

after all, every single self help book out there is written as this vs that so that it gives us the impression of understanding without having even gotten close to scratching the surface.

I truly understand the temptation of explaining an extremely complex topic as a binary choice: evil vs good, red vs blue, pepsi vs coke, ...

let me say it a bit more clearly: This might mean not building agentic systems at all for most of the real use cases most of us have to deal with. ^^ this is what they get right. however, there are so many things I wish they had done differently.

"When building applications with LLMs, we recommend finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all. ... and you should consider when this tradeoff makes sense."

now, the part that most of the folks above ignore or fail to notice is this part of their article very close to the beginning:

I understand and appreciate Anthropic's article as an appropriate reaction to the hysteria of "oh, I have 23 agents running loose, talking to each other, and running my business" --- yea, right!

"I don't want workflows, I want agents" this is a common thing I started hearing from my commercial clients and bootcamp participants (who probably have read the Anthropic article on this topic - or more likely have seen the title as they were scrolling through social media) 1/n

most of us think that the barrier between us and the cool agentic system we want to build (or any system for that matter) is the framework, model, data etc while in reality it is strictly limited by how cleanly we think / problem solve.

one of the commonly quoted productivity hacks is "don't be reactive about what you work on". how have you realized this in your work? esp if your work has lots of dependencies on other people (developers, sales people, customers, etc)

@darrenmarble Darren Awesome about the show - tell us more? Book Launcher is more than a ghostwriter. It's an agentic AI system for launching a book on X. It takes your book content, ideal reader profiles, trends on X, optimizes your hooks and generates drafts for your review. After

with some collaborators we are submitting a proposal in the overlap of RL / LLMs / Economics / Game theory etc do you know any young and ambitious researcher, ideally faculty member, who is very experienced in RL and very interested in LLMs that might want to join the effort?

Washington DC people! I'm hosting an AI pub night on Thursday this week. If you are around and up for some LLM chat over a drink / nibble check this out https://t.co/41SfZLXWYW

maybe controversial but most things people call "agents" aren't really agents (in the sense that they don't even get close to true autonomy - the infrastructure prerequisites for robust performance is just too hard to achieve currently). prove me wrong!

I'll be an instructor at Cooperathon 2024 helping founders leverage AI in their ventures. register here: https://t.co/yAenCD4atS discount code: Cooperathon100 workshops will be in both Eng and Fra! the goal is to walk away from the workshops with prototypes of your ideas

hey people, I'm doing a thing!

2/2 all you need to do is ask yourself: will I realistically miss anything if I continue not seeing these in my inbox esp given that I have barely opened their emails anyway! do your mental health a favor

1/2 hint: end of year is the best time to unsubscribe from all the mailing lists you haven't opened this year! or at least recently. esp since everyone is going to send you "happy holidays" wishes and all those lists that you don't even know you're on would bubble up!