Skip to content
Hands-on AI Playbook

Example: Content Calendar Planning

This page walks through the complete output of running a real workflow through the AI Workflow Framework. The workflow is Content Calendar Planning — a weekly process for planning and sequencing content across LinkedIn, Substack, X, and YouTube.

The first four steps of the framework — Analyze, Deconstruct, Design, and Build — produced three key deliverables for this workflow. Each one is a detailed markdown document. This page summarizes what’s in each and why it matters — then links to the full file on GitHub where you can read every table, every decision point, and every failure mode at full width.

The Deliverables

Section titled “The Deliverables”

1. Workflow Requirements (Step 2 — Deconstruct)

Section titled “1. Workflow Requirements (Step 2 — Deconstruct)”

GitHub View full Workflow Requirements on GitHub

The Workflow Requirements is what Deconstruct produces. What started as “I plan content on Sundays” became 10 refined steps across four phases after the six-question deep dive.

What’s inside:

  • Outcome and Metadata — workflow name, trigger, owner, business objective, definition type
  • Steps Overview — a scannable numbered list of all 10 steps
  • Step Details — each step captured as Goal / Inputs / Outputs / Rules & Edge Cases / Context Needed
  • Sequence — sequential vs. parallel steps, critical path
  • Context Inventory — 9 artifacts the workflow needs, with stable IDs, status (Exists / Needs Creation), AI accessibility, and location
  • Acceptance Criteria — what good output looks like, dimensions that matter, minimum bar
  • Example Scenarios — representative inputs with what to look for in the output
  • Human Gates — where human review is required

Key insight: Notice how each step’s Rules & Edge Cases block captures not just what to do, but what to do when things go wrong. Step 1 has a fallback for when no metrics data exists. Step 2 detects an empty backlog and escalates the importance of Step 3. Step 9 limits refinement to 3 rounds to prevent endless iteration. Capturing edge cases as first-class requirements is what makes the workflow robust enough for AI to execute.

2. Design Spec (Step 3 — Design)

Section titled “2. Design Spec (Step 3 — Design)”

GitHub View full Design Spec on GitHub

The Design Spec is what Design produces from the Workflow Requirements. It classifies each step on the autonomy spectrum, identifies which steps should become reusable skills, and recommends an implementation order.

What’s inside:

  • Orchestration mechanism selection — why “Skill-Powered Prompt” was chosen over a full agent (the creative middle steps benefit from human judgment)
  • Step-by-step decomposition table — each step classified by Phase, Autonomy level, Building Blocks needed, Skill candidacy, and whether it’s a Human Gate
  • Autonomy spectrum summary — Deterministic (4 steps), Guided (4 steps), Human (2 steps)
  • 4 skill candidates with full specifications: purpose, inputs, outputs, decision logic, and failure modes
  • Dependency map and critical path analysis
  • Context inventory — what data sources and reference materials each step needs
  • Recommended implementation order — Quick Wins first (prompt + 2 simple skills), then Guided (2 complex skills), then Future Enhancements

Key insight: The spec doesn’t say “build everything at once.” It recommends starting with a prompt — pure conversation, no infrastructure — so you get value immediately. Database skills layer in incrementally. This build order means you’re running the workflow on day one.

3. Workflow Prompt (Step 4 — Build)

Section titled “3. Workflow Prompt (Step 4 — Build)”

GitHub View full Workflow Prompt on GitHub

The workflow prompt is one of the platform artifacts that Build (Step 4) produces. This is the ready-to-run prompt you paste into any AI tool to execute the workflow.

What’s inside:

  • Full 10-step orchestration — the AI knows exactly what to do at each step, what to present, and when to pause for human input
  • Phase-by-phase instructions — Input Gathering runs skills automatically, Planning proposes and waits for approval, Approval iterates collaboratively, Execution commits to the database
  • Embedded context — channel cadence targets, channel-format fit rules, and content pillar definitions are baked into the prompt so no external documents are needed
  • Input and output specifications — what the user provides (fresh ideas, calendar context, theme selection, approvals) and what the workflow produces (input summary, draft plan, final plan, database confirmation)

Key insight: The prompt reads like a conversation script. At each step, the AI knows whether to act autonomously, propose and wait, or ask for input. The human-in-the-loop gates (Steps 3, 5, 6, 9) are explicit — the AI won’t proceed without approval at those points.

What to Notice

Section titled “What to Notice”

A few things to take away from this example:

  • The expansion. “I plan content on Sundays” became 10 steps across 4 phases, with decision logic, failure modes, and a dependency map. That expansion is what makes the workflow executable by AI.
  • The autonomy spectrum. Not every step needs AI autonomy. Steps 1, 2, 4, and 10 are deterministic (fixed data operations). Steps 5-8 are guided (AI proposes, human decides). Steps 3 and 9 are human-led. The framework helps you see this clearly.
  • The build order. The Design Spec doesn’t just say “build everything.” It recommends starting with a prompt (pure conversation, no infrastructure needed), then layering in skills incrementally. You get value from the first run.
  • Platform-agnostic. The Workflow Requirements and Design Spec work with any AI tool. The skills and MCP connections are implementation details that vary by platform — but the underlying logic is the same everywhere.

Try It on Your Own Workflow

Section titled “Try It on Your Own Workflow”

Ready to run your own workflow through the framework?

Or start from the AI Workflow Framework overview for the full methodology.