Workflow Decomposition Framework
This is the same decomposition logic behind every workflow in today's Toolkit — the thought process that turns "I wish this were automatic" into something you can actually build.
Workflow decomposition is the key to automation
You can't automate a task you can't decompose. Every step has the same anatomy.
-
Break it down first.
If you want to automate a repetitive task, start by breaking those tasks into smaller steps. Smaller than you think — steps that feel obvious are almost always multiple steps when you actually map them out.
-
Every step has 3 parts: an input, a process, and an output.
An input — some data coming in (you may have more than one). A process — what you're asking AI to do with it. An output — what comes out the other side, ready for the next step.
-
Chaining steps together is how you build AI workflows.
The output of step one becomes the input of step two. Because steps are explicit, you get natural checkpoints to debug AI output — you can trace exactly which step went wrong.
-
It's not intuitive because our brains do these things automatically.
We're decomposing work we don't consciously think about. If you don't know where to start, ask AI to help you map out the steps first.
Example: Writing a meeting follow-up
Everyone's sat in a meeting that needed a follow-up (a PTA meeting, a parent-teacher conference, a pediatrician visit — you name it). It's a simple, familiar task — which makes it a good one to practice decomposing on.
What good looks like
A good meeting follow-up captures:
Breaking it down
Here's how to approach this step by step. Click each step to reveal inputs and outputs.
The most important thing is knowing what you would do if you didn't have AI. Where are you pulling information from? What steps do you take? If you don't know the task, AI isn't going to know the task.
AI can be incredibly useful — but only if you can articulate what useful actually looks like for you. That's why someone else's prompt rarely works as well for you as it did for them: it was built around their definition of good, not yours. The real unlock isn't a better prompt — it's taking the time to spell out your own needs.
Choose one useful task from your map. Give AI the inputs and describe the result you want. Test what comes back, improve the instructions, and save what works before adding the next task.
The method we'll practice
- Map the workflow.
Describe what good looks like, then identify the inputs, steps, and outputs. Mark where a person should review or decide.
- Start with one useful task.
Gather what it needs and ask AI to help you build it. For the meal planner, the first goal is a weekly plan that fits your family.
- Test it and improve it.
Try a real example, check the output, and explain what should change. Run it again to see whether the correction worked.
- Save it, then add the next task.
Save the instructions and useful context in your Household OS. Once the meal plan works, it can become the input for a grocery list.
- Decide how to run it again.
Asking in chat is enough. A schedule or other automated trigger is optional, after the workflow works reliably.