Ever watched your AI workflow go off the rails because the LLM decided to "think" instead of just doing the thing? Or gotten stuck somewhere between "cool prototype" and "actually works in production"?

I've been building automation workflows for 15 years. Started with Zapier back when it was new, moved to n8n, and this year I added AI to the mix. Some patterns became obvious pretty quickly. Some old rules still apply. Some new ones are specific to AI.

This talk is field notes, not case studies. I'll share what separates workflows that actually work from expensive disappointments. Process first, subtasks over magic, why your AI is only as good as your underlying system... and a few more things I learned the hard way.

25-30 minutes. Suitable for developers, automation enthusiasts, and anyone building AI-powered workflows. Some familiarity with automation tools helpful but not required.