This page was translated from Dutch with AI.

November 6, 2026by Steff Vanhaverbeke

Let AI prepare the work and an ordinary workflow send it

In a training course on automation, we ran a workflow at a fixed time. A participant asked me if that could also just be done with Claude or ChatGPT. And they suspected that you would use a lot more tokens then. Perhaps you have the same doubt, now that you hear everywhere that agents can do everything. Then building an ordinary workflow seems a bit old-fashioned.

The suspicion was correct. And I would add something else to it: you do not actually have to choose. The setups that run well in practice use both, each for the part they are good at.

Copy the prompt at the bottom of this article. With that, you split your own process into a part for your agent and a part for your workflow.

What comes in and what goes out

I make a simple distinction in my workshops. Everything that comes in is usually variable. Emails from clients, invoices that come as an attachment one time and via a website the next, questions in all shapes. That needs to be read and interpreted, and that is what an agent or a skill is good at.

What goes out is in your own hands. Think of a post on social media, or a report that goes to the same people every week. Or timesheets that you pull from your software and forward. That is always the same. And what is always the same, you actually automate more easily with an ordinary workflow, in n8n or another tool.

The trick is in the folder in between

Suppose you want to post something on social media every day. Then I would say: build an automation in Craft Agents or Claude Code that creates the content and prepares everything neatly. For example in a folder in Google Drive, every post in the same format. And alongside that you have one workflow that checks every day what is ready, and simply posts it. No AI is involved in that anymore.

That fixed structure in the middle actually does the real work. The agent can be creative, as long as it delivers its result in the same place and in the same shape every time. And the workflow does not need to understand anything. It only needs to know where to look.

You can also let your agent do the posting. That does work, but then you use more tokens. It then has to think again every day about something that has already been decided. And the connectors of a tool like n8n are strong, they connect with just about everything. You are then actually paying a thinking model to press a button.

Frequently asked questions

Why should I still build a workflow if an agent can do everything?

An agent can do a lot, but it thinks at every step, and that costs tokens. For what is always the same, like publishing a prepared post, an ordinary workflow does the same work without AI.

What work do I give to an agent and what to a workflow?

What comes in and is different every time, like emails and documents in varying shapes, you give to an agent. What goes out and is always the same, like a fixed publication or a weekly report, you give to a workflow.

How do I make those two work together?

Let the agent prepare its result in the same place and in the same shape every time, for example in a folder in Google Drive. The workflow checks at a fixed time what is ready and handles it.

Where can I learn this hands-on?

In 'AI automation with n8n and MCP' you build workflows yourself and connect them to your agent. For the agent itself, 'Claude AI Power User' is also a good fit. You can find all dates at Workshops & Events.

Keep reading for free

Register free with your e-mail address and read on. Your account opens the free articles in the learning space; the rest is for subscribers, from Learn Plus up.

No comments yetRead the comments

Comments

Members comment here and read each other's comments. Register for free or sign in to join the conversation.

No comments yet. What do you take away?