Your last n8n workflow: how did you start, and how did it go?

Feedback
1

Hi everyone, I’m defining a product and would like to understand how you N8N users actually get from an automation idea to a working n8n workflow. Rather than asking what features you would like, I’d love to hear about the most recent workflow you tried to build — including attempts you abandoned. A short reply to any of these would help:

  1. What were you trying to accomplish, and how did you start? For example: a blank canvas, an official template, a workflow found elsewhere, AI assistance, or a combination.
  2. What made you choose that starting point?
  3. What happened between that starting point and your first useful result? What helped, what required changes, and where—if anywhere—did you get stuck?
  4. Where did you end up? Testing, running regularly, changing approach, or abandoning it?
  5. If AI was involved, please mention which tool you used and what you used it for. If a template was involved, a public link is helpful but optional. A little context would also help: your experience with n8n, Cloud or self-hosted, and whether you were building for yourself, your team, or a client.

Thank you guys! This will help a lot to keep improving how we creare automatons with agents.

2

For my most recent workflow I started from an official template rather than a blank canvas, mostly to learn how the nodes were wired before changing anything. It was a scheduled job that pulled rows from a sheet, filtered them, and sent a formatted summary. What actually consumed the time was not the happy path but the edges: the empty result set, making re-runs safe so a retry did not double-send, and getting the message formatting right. Building the smallest working version first and testing each node with pinned data before running the whole flow helped most. I have abandoned attempts too, usually where I tried to cram classification into one large Code node instead of splitting it into steps with a Set node between them. For your research: templates and AI get people to a first draft fast, but the real sticking point is error handling and idempotency, not the initial build.