Can you clarify "dynamic agentic workflows?" I use their agent node and it works great but what are you trying to do with it? FYI they just added native support for MCP in the latest release both for exposing their workflows as a MCP server and consuming MCP servcies easily via their MCP client tool:
https://docs.n8n.io/integrations/builtin/core-nodes/n8n-node...https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-n...
That should make things interesting because they have a fairly extensive library of templates which can now easily be converted to MCP servers and be more easily invoked agnatically by LLMs.
The "doorway effect" seems to be one expression of that larger phenomenon in which memory recall is largely influenced by the context in which it was encoded. It's the reason that if you're cramming for an exam, to the extent you can recreate the exact conditions of your study environment to mirror those of the test-taking environment, you'll do better.