[threecheese]

Their deployment stuff has been turning me off lately; everyone is rushing to monetize - which I understand and support - but I feel like Langsmith is creeping further and further into Langchain|graph and it makes me hesitant to invest. It’s giving AWS-like gentle but firm lock-in vibes, I wonder if they have any PMs from there.

I do like the way they’ve been able to leverage Langgraph workflows to build agents - it seems like the right abstraction to me - and I also feel their middleware approach is very Django-y which I also like. Are you enjoying their stack?

[solarkraft]

I’m only in the research phase of my hypothetical project so far, so I’m going more off of vibes than personal experience for now.

I’m interested in LangGraph because it seems the closest to an industry standard - every use case seems to be addressed with a tutorial (both first and third party) and there’s an ecosystem of already available graphs/agents. I’m aiming for both high extensibility (new use cases should be easily implementable) and high reliability. The LangGraph docs do a pretty good job at convincing me that they got the latter pretty nailed down. It seems like a hard enough problem to question a new solution on this.

I want to build a (highly reliable & controllable) UI for agents more than I want to build the agents themselves, so my hope is that LangGraph has the biggest ecosystem I can plug into.

They do have some funky lock-in attempts, for instance the LangGraph CLI, which acts as a server for their agent protocol (https://github.com/langchain-ai/agent-protocol), is proprietary. However (and this is what I consider indicative of a strong ecosystem) there’s a free reimplementation named Aegra: https://www.aegra.dev/