People often decide to "do an AI project" when 80% of their problem is solved by a €300 n8n workflow. The reverse is also true: trying to interpret emails with RPA is set up to fail.
| Situation | Traditional (n8n/RPA) | AI agent (LLM) |
|---|---|---|
| Structured data (Excel, API) | ✅ Optimal | ❌ Wasted spend |
| Reading and understanding emails / PDFs | ❌ Brittle | ✅ Strong |
| Clear rules (if X then Y) | ✅ Cheap, fast | ❌ Overkill |
| Natural-language customer interaction | ❌ Frustrating | ✅ Strength |
| Connecting many systems | ✅ Built for it | ⚠️ Still needs n8n |
| Context-dependent decisions | ❌ Cannot | ✅ Made for this |
An n8n workflow is typically €600–2,500 to build with €10–50 / month in usage. An LLM-based agent is €4,500–15,000 to build with €30–300 / month in API costs. That is exactly why you should first ask what can be done with rules.
Well-built solutions have both an LLM and n8n in them. n8n connects systems and handles triggers and execution; the LLM handles the "smart steps" — categorisation, extraction, response generation. They are not rivals; they complement each other.
Ask yourself: "Could I write the rules on one page so that a junior colleague would execute them flawlessly?" If yes → n8n / RPA. If no ("you have to feel it" type decision) → LLM.
Not sure what you need? A 30-minute call usually settles it.