Module 03 · Tier 4 — Intermediate Prompting

Designing AI Workflows

Map your tasks. Identify where AI adds leverage. Build repeatable systems, not one-off experiments.

13 min Tier 4 — Intermediate Prompting

Why This Matters

A prompt is a single move. A workflow is a system. The difference matters because systems scale and compound in ways that individual prompts don't. Once you've designed a workflow — mapped the task, identified where AI adds leverage, defined the handoffs — you can run it repeatedly, hand it to others, and improve it over time. This module is about thinking in workflows, not just prompts.

The Concept

What Makes Something a Workflow vs. a Prompt

A workflow has:

A defined input (what triggers it)
A clear sequence of steps (including which steps are AI-assisted and which aren't)
A defined output (what it produces)
Quality criteria (how you know it worked)

A prompt has none of these. It's a one-off request. Workflows are the difference between using AI occasionally and using it systematically.

The Task Mapping Process

Before you can design a workflow, you need to understand the task at the component level. Most tasks that seem like single things are actually sequences of 4-8 subtasks. Breaking them down reveals where AI adds the most leverage and where human judgment is essential.

The mapping questions:

What are the individual steps in this task, from start to finish?
For each step: is this primarily information processing, generation, or judgment?
Which steps have high AI leverage (generation, synthesis, formatting)?
Which steps require human judgment that AI shouldn't replace?
What are the quality checkpoints — where must a human review before proceeding?

The AI Leverage Matrix

Not all task components benefit equally from AI. A useful mental model:

High AI leverage (automate or heavily assist): First drafts, research synthesis, formatting and reformatting, generating options, translating between styles/audiences, summarizing large inputs.

Medium AI leverage (use AI, then review carefully): Editing and refinement, analysis with provided data, argument structuring, identifying gaps.

Low AI leverage (AI input only, human decides): Final judgment calls, relationship-sensitive communication, decisions with ethical implications, anything where being wrong has significant consequences.

Workflow Design Patterns

Pattern 1: Draft → Refine → Polish
AI generates a first draft. Human evaluates and directs revisions. AI polishes. Human approves. Good for: writing, reports, proposals.

Pattern 2: Research → Synthesize → Apply
Human provides source material. AI synthesizes and extracts. Human applies judgment to synthesized output. Good for: research tasks, competitive analysis, literature review.

Pattern 3: Options → Evaluate → Develop
AI generates multiple options. Human evaluates and selects. AI develops the chosen option. Good for: strategy, creative work, problem-solving.

Pattern 4: Input → Transform → Verify
Human provides structured input (data, notes). AI transforms into desired format. Human verifies accuracy. Good for: data formatting, document generation, template population.

Documenting Your Workflows

The single most valuable thing you can do after building a workflow that works: write it down. A documented workflow has:

Trigger: what situation calls for this workflow?
Steps: each step, with the prompt for AI steps
Handoffs: where human review happens and what to check
Output: what the workflow produces
Pitfalls: what went wrong the first few times and how to avoid it

A one-page workflow document takes 10 minutes to write and can save hours over the following months.

Mapping a real workflow

Task: Prepare for a client meeting about a new project proposal.

Typical approach (no workflow): Review some notes, maybe look up the client, wing it.

Mapped workflow:

Input: client name, project brief, previous interactions — Human provides
Research synthesis: summarize what AI knows about the client's industry challenges — AI, high leverage
Gap identification: what do I not know that I should? — AI generates questions, human judges which matter
Objection prep: likely concerns about this type of proposal — AI generates, human refines
Opening framing: draft a meeting opening that acknowledges their context — AI drafts, human edits
Final review: is this accurate and appropriate for this specific relationship? — Human only

Total AI time: 15 minutes. Result: a meeting preparation that would have taken 45 minutes without AI — and would have been less thorough.

Hands-On Exercise

Map and document one workflow

ClaudeChatGPTNotionAny notes tool

Choose a task you do regularly — ideally one that takes 30+ minutes and involves both information processing and judgment. Step 1: Write out every substep (aim for 6-10 steps). Step 2: For each step, mark it: High AI leverage / Medium AI leverage / Low AI leverage. Step 3: Identify which pattern it fits (Draft→Refine→Polish, Research→Synthesize→Apply, Options→Evaluate→Develop, or Input→Transform→Verify). Step 4: Run the workflow once using AI for the high-leverage steps. Step 5: Document it: trigger, steps with prompts, handoffs, output, pitfalls. Time how long it takes with versus without AI. That number is your ROI estimate.

Don't try to automate everything in the first pass. High-leverage steps first, human judgment steps always.

Active Recall

Before moving on — close this lesson and answer these from memory. Then come back and check. Testing yourself (not re-reading) is how this sticks.

1 What are the four components that distinguish a workflow from a prompt?

2 Describe the AI leverage matrix. Give one example of a task component that belongs in each of the three categories.

Reflection

What is the one task in your work that would benefit most from being turned into a documented workflow? What's preventing you from designing it today?

Key Takeaway

Workflows compound; prompts don't. Map tasks to components, identify where AI adds leverage, design the handoffs where human judgment applies, then document the workflow so it's repeatable. One well-designed workflow is worth more than a hundred individual prompt experiments.