A workflow is the whole job
“Write an email” is a request. “Take the notes from a client call, draft a recap, ask me to check it and save the approved version” is a workflow.
The difference is that you have described where the work starts and where it ends.
You can run that sequence manually. Automation comes later, when a trigger starts some or all of the steps for you.
To see one end to end, get the free AI business starter.
What belongs in the workflow?
- Input: the material the AI needs, not every file you own.
- Instructions: what to produce and what a good result looks like.
- Review: the details you want checked before the next step.
- Destination: where the result belongs.
- Exception: what happens when something is missing.
Example: after a client call
A designer copies the call notes into an AI tool. The tool separates decisions from ideas, drafts a friendly recap and lists unanswered questions.
The designer checks the names, dates and commitments before sending. The approved recap is saved beside the project notes.
If the notes contain no delivery date, the useful response is a question—not a confident invented deadline. That is why the exception belongs in the workflow.
AI workflow vs AI agent
In Anthropic’s engineering terminology, a workflow follows a predefined path; an agent can decide which steps or tools to use as it works toward a goal.
A fixed notes-to-recap sequence is a workflow. An agent researching a question may choose which sources to open next.
Products use these labels differently, so look at the behaviour rather than the name. More on agents: what an AI agent is.
When should you automate it?
Run the process yourself first. Try a clean example, a messy example and one with missing details.
When you know what good looks like, decide which repetitive step to connect.
Perhaps new notes should start a draft automatically while sending remains your decision.
Measure the complete job, including checking and corrections. A fast draft that creates more cleanup is not a better workflow.
What is a good first workflow?
Choose something you already do often: a meeting recap, a weekly project summary or turning a transcript into a content outline.
Familiar work gives you a practical test. You can tell whether the result helps without becoming an AI expert first.
Read how to use AI in your business for a step-by-step starting project.
Technical distinction: Anthropic: Building effective agents.