I’ve seen some info talking about Langchain in the official docs, but not sure if the current AI Agent nodes use Langchain or a completely different system? If it’s a different system, how it compares to Langchain (in terms of flexibility, results, etc.)?
Also, I’ve seen in a few places that they say that n8n agents are quite limited (like just some basic AI brain in the middle of a traditional workflow), not proper/advanced AI agents with advanced memory, multi-step reasoning, custom data stores, orchestration, etc. Is it true?
Obviously I don’t expect n8n agents to be the most sophisticated in the market, but would love to have more info (and real experiences) to understand if they are real agents that can manage some long/complex tasks by themselves or only useful/reliable for very simple tasks inside a workflow.
Are n8n agents similar at least to other no-code/low-code AI agents (CrewAI, Lindy, Relevance, Flowise, etc.) or their agents are clearly superior?
n8n AI Agents don’t use LangChain.
They run on n8n’s own internal agent system that focuses on tool-calling and workflow orchestration.
They’re good for:
multi-step reasoning
using n8n nodes as tools
short-term memory within a workflow
automating structured tasks with AI
They’re not as advanced as LangChain/CrewAI for:
long-term memory
complex planning
multi-agent systems
full autonomous behaviour
So they’re more powerful than simple no-code agents (Flowise), but less advanced than LangChain/LangGraph or CrewAI. They are designed for practical, reliable workflow automation, not for fully autonomous agents.
I think I will add another topic for this related question, but just in case you can help me:
Do you still use “normal” AI nodes or only AI agents?
If you want to include an AI step in a workflow that is not very complex, do you use the normal AI nodes (an OpenAI “message a model” node for example) or do you use AI agent nodes for all AI tasks now?
Pros and cons of both options?
Some tasks where you will always use/recommend one of the options?
Recently I’ve tried n8n AI workflow builder. I asked it to include a simple AI step using OpenAI API and it created an agent (not a normal OpenAI node). Not sure if it’s the AI builder doing “strange” stuff or using agents for most AI tasks is now the best practice.