As a hack, I have got LangGraph working inside of a node-red flow. I will write it up as a medium article in the next few days, as it might be of help to others. It's very codey.
I think production engineers will have to master the Langchain ecosystem, when we creating mission critical AI based systems. I think node-red has good nodes in place to record decisions and data as part of an AI ethics program.
Flowise have a good set of nodes around langchain, but I bet it was a massive and expensive undertaking to build.
Also Dashboard 2 is something n8n is sorely missing and is needed to monotor live systems.
However, there is something you mention that I think is important - dashboard 2 is still missing critical functionality—for instance, I cannot set it up as the “/“ (homepage) URL in an instance (there’s a lingering bug for that for around a year now) and it is still far from being as well integrated as the original.
The bit I want to surface in this thread is that there is actually a lot that can be done with sub-flows and custom functions alone (even without added third-party modules) - I built my own library of AI/“agentic” subflows with raw HTTP request nodes and a simple convention for passing LLM context in payloads so I’m a bit biased here, but overall I find Node-RED superior in the amount of control I have over how data is handled.
(Also, I am running Node-RED in bun rather than NodeJS and it is way more efficient, although that means I can’t rely on some third-party nodes, but that is another story)
As ever, I think each has its strengths and weaknesses, but I know which one I prefer...
In terms of 'out of the box' experience, I do agree there is space for some improved Node-RED nodes to support some of these workflows. There certainly are nodes in the community to help, but there's space for more. And that's an area I think gives Node-RED an advantage; the open source nature that allows anyone to build in the gaps of what's already there. Node-RED provides plenty of low-level building blocks to connect with pretty much anything - but having higher level abstraction nodes is needed to make it more accessible.
It's also an area we're looking at in my day job (FlowFuse CTO) where we provide a commercial platform built around Node-RED.