Day 4: What Makes a Workflow "AI-Powered?"
7-Day email course, workbook and ultimate guide
Yesterday we broke down the anatomy of a workflow.
Trigger → Steps → Output.
Simple structure, infinite applications.
Today we add the AI layer and this is where things get interesting.
Traditional automation vs AI automation
Traditional automation is good at repetitive, predictable tasks. Move this file here. Send this email when that happens. Update this spreadsheet with that data.
It follows rules. If X, do Y. Every time, the same way.
But some tasks need more than rules. They need judgement.
That’s where AI comes in.
Quick interlude: When I say AI throughout this course. I mean Generative AI and Large Language Models (LLM’s) within ChatGPT, Claude, Gemini etc.
What AI can do that regular automation can’t
AI can handle tasks that require understanding, interpretation, or creativity.
Things like:
Reading and understanding: Making sense of an email, a document, or a transcript
Summarising: Condensing a long piece of content into key points
Categorising: Deciding what type of thing something is (complaint vs question, urgent vs not urgent)
Extracting: Pulling specific information out of unstructured text
Drafting: Writing content based on inputs (emails, summaries, reports)
Deciding: Making judgement calls based on context
These are the “thinking” tasks. The ones that used to require a human brain. In AI circles, it’s called reasoning.
Where AI fits in a workflow
AI doesn’t replace the whole workflow. It handles specific steps within it.
Look at a workflow and ask: which steps require understanding or judgement?
Those are your AI steps. Everything else can be traditional automation.
Example: Meeting follow-up workflow
Here’s a simple workflow that combines both - one you may be familiar with.
Trigger: Meeting ends (calendar event)
Steps:
Get the meeting transcript (automation)
Summarise the discussion into key points (AI)
Extract action items with owners and deadlines (AI)
Draft a follow-up email (AI)
Send the email to attendees (automation)
Create tasks in project tool for each action item (automation)
Log the meeting in the tracker (automation)
Steps 2, 3, and 4 need AI. They require understanding what was said and making decisions about what matters.
Steps 1, 5, 6, and 7 are pure automation. They’re just moving data and triggering actions.
When you need AI vs when you don’t
Not every workflow needs AI. Sometimes simple automation is enough.
Use AI when:
The input is unstructured (emails, documents, transcripts)
The task requires interpretation or judgement
The output needs to sound human
Rules alone can’t handle the variation
Use simple automation when:
The input is structured (form fields, database records)
The task follows clear, consistent rules
You’re just moving data between systems
The same action happens every time
Adding AI when you don’t need it just adds complexity and cost. Keep it simple where you can.
The important bit
AI is a tool, not magic. It’s very good at certain things and not so good at others.
Your job isn’t to use AI everywhere. It’s to know when AI adds value and when it doesn’t.
That judgement? That’s the human bit. And it’s not going anywhere.
Tomorrow, we’ll talk about how to spot automation opportunities in your own work. You’ll start seeing them everywhere.
Speak then,
Tim
Your course workbook
I’ve built a guide (in Notion) to go alongside these emails.
Inside you’ll find:
- Daily AI prompts to reinforce each lesson.
- Exercises to spot opportunities in your own work.
- Space to capture automation ideas as they come to you.
Complete the exercises, and you’ll be ready for the full guide on Day 7.
Sign up for Notion (it’s free and it’s awesome).
Then get your course workbook.