Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Why we no longer use LangChain for building our AI agents

octomind.dev

Post reply on HN
Print view View original

91–100 of 307 posts

Re: Why we no longer use LangChain for building our AI agents

#91

Post by CGamesPlay »

wg0 wrote: post #89

Sorry noob question - where can I read more about this "agents" paradigm? Is one agent's output directly calling/invoking another agent? Or there's already fixed graph of information flow with each agent (I presume some prompt presets/templates like "you are an expert this only respond in that") sorts of? Also, how much success people have or had with automating the E2E tests for their various apps by stringing such…

Fundamentally, "Agent" refers to anything that operates in an "observe-act" loop. So in the context of LLMs, an agent sees an observation (like the code base and test output) and produces an action (like a patch), and repeats.

Re: Why we no longer use LangChain for building our AI agents

#92

Post by isaacfung »

muzani wrote: post #68

Langchain was released in October 2022. ChatGPT was released in November 2022. Langchain was before chat models were invented. It let us turn these one-shot APIs into Markov chains. ChatGPT came in and made us realize we didn't want Markov chains; a conversational structure worked just as well. After ChatGPT and GPT 3.5, there were no more non-chat models in the LLM world. Chat models worked great for everything, inc…

I am not sure what you mean by "turn these one-shot APIs into Markov chains." To me, langchain was mostly marketed as a framework that makes RAG easy by providing integration with all kinds of data sources(vector db, pdf, sql db, web search, etc). Also older models(including initial chatgpt) had limited context lengths. Langchain helped you to manage the conversation memory by splitting it up and storing the pieces in a vector db. Another thing langchain did was implementing the react framework(which you can implement with a few lines of code) to help you answer multi hop problems.

Re: Why we no longer use LangChain for building our AI agents

#93

Post by mstipetic »

Kiro wrote: post #8
CharlieDigital wrote: post #2

Bigger problem might be using agents in the first place. We did some testing with agents for content generation (e.g. "authoring" agent, "researcher" agent, "editor" agent) and found that it was easier to just write it as 3 sequential prompts with an explicit control loop. It's easier to debug, monitor, and control the output flow this way. But we still use Semantic Kernel[0] because the lowest level abstractions tha…

What's the difference? I thought "agents" was just a fancier word for sequential prompts.

Sequential prompts with an occasional cron job

Re: Why we no longer use LangChain for building our AI agents

#94

Post by zEddSH »

wg0 wrote: post #89

Sorry noob question - where can I read more about this "agents" paradigm? Is one agent's output directly calling/invoking another agent? Or there's already fixed graph of information flow with each agent (I presume some prompt presets/templates like "you are an expert this only respond in that") sorts of? Also, how much success people have or had with automating the E2E tests for their various apps by stringing such…

> Also, how much success people have or had with automating the E2E tests for their various apps by stringing such agents themselves together?

There’s a few startups in the space doing this like QA Tech in Stockholm, and others even in YC (but I forgot the name). I’m skeptical of how successful they’ll be, not just from complex test cases but things like data management and mistakingly affecting other tests. Interesting to follow just in case though, E2E is a pain!

Re: Why we no longer use LangChain for building our AI agents

#95

Post by pavi2410 »

wg0 wrote: post #89

Sorry noob question - where can I read more about this "agents" paradigm? Is one agent's output directly calling/invoking another agent? Or there's already fixed graph of information flow with each agent (I presume some prompt presets/templates like "you are an expert this only respond in that") sorts of? Also, how much success people have or had with automating the E2E tests for their various apps by stringing such…

I want to learn about agents too!

Re: Why we no longer use LangChain for building our AI agents

#96

Post by hcks »

wg0 wrote: post #89

Sorry noob question - where can I read more about this "agents" paradigm? Is one agent's output directly calling/invoking another agent? Or there's already fixed graph of information flow with each agent (I presume some prompt presets/templates like "you are an expert this only respond in that") sorts of? Also, how much success people have or had with automating the E2E tests for their various apps by stringing such…

Don’t waste your time, it’s been around since GPT3, and had no results so far. Also notice how no frontier lab is working on it.

Re: Why we no longer use LangChain for building our AI agents

#97

Post by zby »

wg0 wrote: post #89

Sorry noob question - where can I read more about this "agents" paradigm? Is one agent's output directly calling/invoking another agent? Or there's already fixed graph of information flow with each agent (I presume some prompt presets/templates like "you are an expert this only respond in that") sorts of? Also, how much success people have or had with automating the E2E tests for their various apps by stringing such…

In practice this means function calling - the LLM chooses the function to call (and its parameters). Usually in a loop with a 'finish' function that returns the control to the outside code.

You can do that without function calling - as did the original ReAct paper - but then you have to write your own grammar for the communication with the LLM, a parser for it, and also you need to teach the LLM to use that grammar. This is very time consuming.

Re: Why we no longer use LangChain for building our AI agents

#98

Post by zby »

hcks wrote: post #96
wg0 wrote: post #89

Sorry noob question - where can I read more about this "agents" paradigm? Is one agent's output directly calling/invoking another agent? Or there's already fixed graph of information flow with each agent (I presume some prompt presets/templates like "you are an expert this only respond in that") sorts of? Also, how much success people have or had with automating the E2E tests for their various apps by stringing such…

Don’t waste your time, it’s been around since GPT3, and had no results so far. Also notice how no frontier lab is working on it.

Letting the LLM to decide what to do is a powerful technique. For example one pass RAG is very limited: https://zzbbyy.substack.com/p/why-iterative-thinking-is-cruc... To make it iterative you need the cede the control to the LLM.

Re: Why we no longer use LangChain for building our AI agents

#99

Post by weinzierl »

isaacfung wrote: post #92
muzani wrote: post #68

Langchain was released in October 2022. ChatGPT was released in November 2022. Langchain was before chat models were invented. It let us turn these one-shot APIs into Markov chains. ChatGPT came in and made us realize we didn't want Markov chains; a conversational structure worked just as well. After ChatGPT and GPT 3.5, there were no more non-chat models in the LLM world. Chat models worked great for everything, inc…

I am not sure what you mean by "turn these one-shot APIs into Markov chains." To me, langchain was mostly marketed as a framework that makes RAG easy by providing integration with all kinds of data sources(vector db, pdf, sql db, web search, etc). Also older models(including initial chatgpt) had limited context lengths. Langchain helped you to manage the conversation memory by splitting it up and storing the pieces i…

I too wondered about "by "turn these one-shot APIs into Markov chains.".

Re: Why we no longer use LangChain for building our AI agents

#100

Post by geuis »

I built my first commercial LLM agent back in October/November last year. As a newcomer to the LLM space, every tutorial and youtube video was about using LangChain. But something about the project had that "bad code" smell about it.

I was fortunate in that the person I was building the project for was able to introduce me to a few other people more experienced with the entire nascent LLM agent field and both of them strongly steered me away from LangChain.

Avoiding going down that minefield ridden path really helped me out early on, and instead I focused more on learning how to build agents "from scratch" more or less. That gave me a much better handle on how to interact with agents and has led me more into learning how to run the various models independently of the API providers and get more productive results.

Post reply on HN