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

11–20 of 307 posts

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

#11

Post by ec109685 »

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.

Some folks try to orchestrate the whole operation by a higher level prompt that essentially uses function calls to more specific prompts.

Versus just using the LLM’s for specific tasks and heuristics / own code for the orchestration.

But I agree there is a lot of anthropomorphizing that over states current model capabilities and just confuses things in general.

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

#12

Post by dcole2929 »

I've seen a lot of stuff recently about how LangChain and other frameworks for AI/LLM are terrible and we shouldn't use them and I can't help but think that people are missing the point. If you need strong customization or flexibility frameworks of any kind are almost always the wrong choice, whether you're building a website or an AI agent. That's kind of the whole point of a framework. Opinionated workflows that enable a specific kind of application. Ideally the goal is to cover 80% of the cases and provide escape hatches to handle the other 20% until you can successfully cover those too.

As someone new to the space I have zero opinions of whether LangChain is better than writing it all yourself, but I can certainly say that, I at least, appreciate having a proscribed way of doing things, and I'm okay with the idea that I may get to a place where it no longer serves my needs. It's also worth noting that the benefit of LangChain is the ability to "chain" together these various AI links. Is there a better easier way to do that? Probably, but LangChain removes that overhead.

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

#13

Post by CharlieDigital »

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.

Typically, the term "agents" implies some autonomous collaboration. In an agent workflow, the flow itself is non-deterministic. One agent can work with another agent and keep cycling between themselves until an output is resolved that meets some criteria. An agent itself is also typically evaluating the terminal condition for the workflow.

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

#14

Post by lolinder »

etse wrote: post #10

My reading of the article is that because LangChain is abstracted poorly, frameworks should not be used, but that seems a bit far. my experience is that Python has a frustrating developer experience for production services. So I would prefer a framework with better abstractions and a solid production language (performance and safety), over no framework and Python (if those were options)

For most of what people are doing with AI you don't need Python because you don't need the ML ecosystem. You're either going to be talking to some provider's API (in which case there are wrappers aplenty and even if there weren't their APIs are simple and trivial to wrap yourself) or you're going to self-host a model somewhere, in which case you can use something like ollama to give yourself an easy API to code against.

All of the logic of stringing prompts and outputs together can easily happen in basically any programming language with maybe a tiny bespoke framework customized to your needs.

Calling these things "AI agents" makes them sound both cooler and more complicated than they actually are or need to be. It's all just taking the output from one black box and sticking it into the input of another, the same kind of work frontline programmers have been doing for decades.

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

#19

Post by OutOfHere »

altdataseller wrote: post #17

Langchain reminds me of GraphQL. A technology that a lot of ppl seem to hype about, sounds like something you should use because all the cool kids use it, but at the end of the day just makes things unncessarily complicated.

GraphQL actually holds value in my view as it gives custom SQL-like functionality instead of basic JSON APIs. With it, you can do fewer calls and retrieve only the attributes you need. Granted, if SQL were directly an API, then GraphQL wouldn't hold too much value.

Langchain has no such benefit.

Post reply on HN