The Three Roles AI Plays in a Workflow

AI in workflows does three things. Not always in isolation—often all three in one complex workflow—but these are the main patterns:

1. Generator: AI Creates New Content

The workflow gives AI a prompt or a template, and AI writes something new.

Examples: - "When a customer fills out our feedback form, have AI generate a personalized thank-you email based on what they said" - "When I save a voice memo to a folder, have AI transcribe it and turn the transcript into meeting notes" - "When a new article is published, have AI write three social media captions for different platforms"

The data flowing in is usually a brief, a template, or raw material. The data flowing out is new content.

Generator AI is useful because it saves you from blank-page paralysis and takes repetition out of writing. But notice: you're still giving it a spec. A generator doesn't decide what to write. You do. It just does the writing.

2. Transformer: AI Reformats, Rewrites, or Classifies

The workflow gives AI some text or data, and AI changes it in a structured way.

Examples: - "When a customer support email arrives, classify it as 'billing question,' 'technical issue,' or 'complaint,' then route it to the right team" - "When I save a document, extract all the dates mentioned and reformat them as YYYY-MM-DD" - "When a new product review comes in, have AI rewrite it in the tone of our brand" - "When a job application arrives, have AI score it 1-10 based on our criteria and list the top 5 reasons"

The data in is usually a document, email, or unstructured text. The data out is structured, categorized, or reformatted. Transformer AI is powerful because it takes messy human input and makes it organized and consistent.

3. Decision-Maker: AI Routes or Filters Based on Content

The workflow asks AI a yes/no question or "which option" question, and branches based on the answer.

Examples: - "When an email arrives, ask AI: 'Is this urgent?' Route to my urgent inbox if yes, to a daily digest if no" - "When a lead fills out a form, ask AI: 'Based on their company size and budget, are they enterprise or small business?' Route them to the right sales team" - "When feedback arrives, ask AI: 'Is this complaint likely to escalate to social media?' Flag it for priority review if yes"

The data in is usually context (email, form, feedback). The data out is a decision that branches the workflow.

Decision-making AI is valuable, but be careful here: you're asking AI to make judgments that might need human review. We'll talk about that in Lesson 5.