Workflow Assistant

Questions
1

What’s everybody using in terms of a workflow assistant these days?
I tried N8Nchat but it was throwing me errors.

Claude Sonnet 3.7 can read screenshots, which is very helpful. But it’s a couple of years out of date as far as N8N. Many of the node names are different and the screens inside the nodes had different options in 2023. So it’s a lot of back and forth and slow for that reason.

Grok 3 beta has more recent N8N info, but doesn’t seem to be able to read from images.

I’m a relative n00b. Any suggestions for AI assistance for building workflows?

I’m using N8N version 1.88.0 and am self-hosting on Railway.

2

Hi @KeithL
Welcome to the community.
Maybe it is best to start with the n8n courses to get familiar with n8n :wink:
Relying on “AI” sound like a very bad idea to me. Of course when you are familiar with n8n, then you can use it to build flows faster, but just starting with generated workflows will only lead to troubles.

1 Like
3

I guess I’m wrong about Grok. I was accessing Grok via openrouter, and images aren’t possible there.

4

I love n8n: creating workflows, community nodes, and applying hacks.

But be clear. People want results in no time/effort.
That’s why n8n official directory templates makes sense to boost productivity.

Building workflows manually makes sense now. But AI will replace this task almost completely.

Human in the loop will be necessary only to check that any generated workflow with AI satisfies any requirements, and will apply minor changes when required.

5

I haven’t had much luck with the templates. But now that I’ve spent the last few days building workflows, maybe they’ll make more sense to me.

1 Like
6

Sure… The problem is that templates don’t match exactly your requirements.

But this can be easily fixed with a bit of IA and imagination.
We are working on that at n8nhackers.com :wink:

7

And back to my original question – which AI do you prefer for fixing workflows?

8

General purpose AIs (like GPT, Claude…) are focused on replying lots of questions (not only related to n8n). They would be able to fix basic errors when models were generated (one or two years ago).

That’s why most of the AIs fail. N8n is in constant evolution and specs change time to time.

But complex ones need custom training.