Tried talking to the reps about the AI features. All were very cagey, suggesting that they didn't really know and were parroting buzzwordy benefits because the higher-ups told them to. So far, the "AI revolution" seems to have just led to higher price tiers.
This feels very crypto-like in this regard for me. Businesses seem to be struggling with what we're all struggling with, LLMs are untrustworthy.
AI isn't moving the needle much here, which is why GenAI "solutions" are underwhelming things like support chatbots, writing assistance and auto-generated stock art for Facebook ads.
There's nothing wrong with doing that a service but it falls well short of the fantastical visions that Microsoft, OAI and SV at large have portrayed.
Things have cooled a bit more nowadays as the cost economics became more real. I wouldn’t expect another LangChain level fund raise.
Currently I don't find LLMs tremendously useful because I have to be extremely specific with what I want, which IMO is closer to writing code than the magical promise of turning ideas into products.
An app that I can just talk into and say, "I want to do X" and it can start building X while asking me questions for functional requirements, edge cases, UI/UX, that's a killer app.
It's also an app which could actually decimate many SWE jobs, so, I should be careful what I wish for.
I think you are right -- eventually we will see an ecosystem pop up that uses LLMs in unique ways that are only possible with LLMs.
But in the meantime people are using LLMs to shortcut what used to be long processes done by humans, and frankly, there is a lot of value in that.
Even this is IMO too ambitious (due to edge cases, and multi-shot prompting being basically required, which kind of defeats the purpose). Right now, I'm working on a product that can just do simple OS stuff (and even that is quite challenging); e.g. "replace all instances of `David` with `Bob` in all text files in this folder".
The same is not true of AI projects which require a lot more upfront investment.
LLMs are like touch screens: technically interesting and with great upside potential. But until the iPhone, multi-touch, the app ecosystem comes along, they'll remain an interesting technological advancement (nothing more, nothing less).
What I'm also noticing is that very little effort (and money) is actually spent on value generation, and most is spent on pure bloat. LangChain raising $25m for what is essentially a library for string manipulation is crazy. (N.B. I'm not solely calling out LangChain here, there are dozens of startups that have raised hundreds of millions on what is essentially AI tooling/bloat.) We need apps--real, actual apps--that leverage LLMs where it matters: natural-language interaction. Not pipelines upon pipelines upon pipelines or larger and larger models that are fun to chat with but actually do basically nothing.