Hi everyone,
I’ve been using n8n for quite some time now, and lately, I’ve been reflecting on the gap between “it works in the editor” and “it works in production.”
As a founder of an automation agency and one of the first University Lecturers in AI in Poland, I find myself in a unique position where I have to translate complex AI theory into stable, reliable workflows for my clients. After consulting for over 70+ startups, I’ve noticed a recurring pattern: most workflows fail not because of node errors, but because of a lack of architectural resilience.
In my lectures and my agency practice (AI Reveo), I’ve started focusing more on “Negative Path” design:
-
Decoupling Logic: Using PostgreSQL/Supabase as a state machine instead of relying on node memory.
-
Standardized Error Handling: Building global error triggers that don’t just notify, but actually attempt to self-heal.
-
Documentation as Code: Ensuring that every workflow is readable for the person who will maintain it at 2 AM.
I’m curious to hear from this community—at what point do you decide a workflow is “production-ready”? Do you have a specific checklist for error handling before handing it over to a client?
I’m looking forward to connecting with more of you and sharing insights from the intersection of AI academia and practical automation.
Let’s connect on LinkedIn: @robhaluza
Best, Robert