I’m far from technical, but I’ve been experimenting with n8n AI workflows for a couple months.
My most frequent use cases involve aggregating articles from RSS feeds and passing them through AI Agents to filter for relevance and to group by topic/country. A maddening experience is that I can run the same workflow in succession and get dramatically different output each time. Certain articles included in one instance but excluded the next…certain articles grouped correctly and then grouped incorrectly the next time. I typically use Gemini 2.0 flash, but I’ve experienced this inconsistency with other models.
I think my prompts are pretty clear, but in any case shouldn’t the output be quite consistent when the setup is identical? If anyone has advice to solve for this, I’d be very interested.