📄 From the workflow on paper to an executable workflow
The workflow you designed in module 5.1 is already a system. Now let’s translate each box into a prompt and run the sequence in a chat session. No installation. No configuration.
🧩 How it works in practice
In the same chat session: you send Prompt 1, receive the output, and use that output as part of Prompt 2 (“Based on what you just generated, now...”). The AI keeps the conversation context — that’s chaining.
Each box in the workflow became a message in the chat. The sequence of messages is the system.
💡 Practical Tip
To preserve context between boxes, start each subsequent prompt with: "Based on the previous [structure/analysis/draft]..." This activates the AI’s context window and ensures that the output of step 1 becomes the foundation for step 2.
🔗 Prompt chaining — complete 4-step example
The most important technique in the module. Each step inherits the previous one’s work and transforms it. See a complete example of producing an article in 4 linked stages.
Step 1 — Brief and structure
Output: outline with section headings + 1 sentence describing the content of each one. This is the input for Step 2.
Step 2 — First draft
Output: full article in draft. The instruction "don't review yet" is intentional — review is the next step.
Step 3 — Evaluate against criteria
Output: structured diagnosis. If everything is "yes" → go straight to Step 4. If there's a "no" → Step 4 fixes only what failed.
Step 4 — Refined final version
Output: final article, ready for minimal human review and publication. From the brief to here: 4 messages in one session.
Inheritance
Each step builds on the previous one — it doesn’t start from scratch
Separation of roles
Generation and review in separate prompts = higher quality
Explicit evaluation
Criteria written beforehand — not a judgment afterward
Surgical
Fix only what failed — preserve what was good
🗃️ Knowledge bases as part of the system
The system doesn’t start from scratch. The assets you built in the previous learning paths are pieces that fit into the workflow. Injecting them as the first input is what sets your system apart from a generic one.
HOW TO INJECT THE KNOWLEDGE BASE — PROMPT 0 STRUCTURE
Project: [project name — from T4]
Voice/tone: [from the T3 base brief]
Audience: [ICP/profile — from T3]
Approved templates: [from the T1 library]
Quality criteria: [from the T4 rubric]
Constraints: [from T3]
# Confirm you understand the context before I start the system.
How the pyramid connects here
- T1 — Ready-made prompts become system boxes
- T3 — The base briefing becomes the system’s input
- T4 — Playbooks and rubrics become evaluation criteria
- T5 (now) — Connects everything in a flow with a loop
- T6 (next) — The system learns from each cycle
Why inject context at the beginning
- • Avoid re-explaining who you are and what you need in every session
- • Ensure every box in the system works from the same assumptions
- • Turns AI from a question-answerer into a process collaborator
🔄 Review loop — the AI evaluates its own result
A review loop is a prompt that asks the AI: "does what you just generated meet criteria X, Y, Z?" The answer to this question guides the system: if yes, deliver; if no, adjust and repeat.
EVALUATION PROMPT TEMPLATE (the system’s “reviewer”)
1. [Criterion 1]: does it meet it? yes / no → what was missing
2. [Criterion 2]: does it meet it? yes / no → what was missing
3. [Criterion 3]: does it meet it? yes / no → what was missing
If all criteria = yes:
Deliver the final result with one line: "✓ All criteria met."
If any criterion = no:
Rewrite only what didn’t pass. Don’t change what was already correct.
✓ When the loop works well
- ✓Objective, verifiable criteria ("≤ 150 words", "includes a CTA")
- ✓Rubric defined before the system runs, not made up afterward
- ✓Maximum 3-4 criteria per loop (too many criteria = paralysis)
✗ When the loop fails
- ✗Vague criterion: "be creative" → AI will always say it met the requirement
- ✗Contradictory criterion: the loop runs indefinitely
- ✗No completion criteria: the system never delivers
✅ Quality criteria built into the workflow
A quality criterion isn’t what you thinks that feels good after you see the result. It’s what you written before the system runs. This distinction separates systems that converge from systems that spin in an infinite loop.
EXAMPLES OF OBJECTIVE VS. VAGUE CRITERIA
✗ VAGUE (system doesn't converge)
"Appropriate tone"
"Good structure"
"Quality writing"
✓ OBJECTIVE (system converges)
"Max. 150 words, 3 paragraphs"
"First sentence: question or statistic"
"End with 1 concrete action"
📊 The 3 types of criteria every system needs
- Format criterion: size, structure, number of sections, file type
- Content criterion: required elements, what must not appear, necessary references
- Tone/voice criterion: formality level, terms to use/avoid, target audience for the text
💡 Practical Tip
Take a result you approved in the past and ask: "why did I approve this?" The answers are your criteria. “I approved it because it was short, direct, had an example, and ended with a next step”—those are 4 objective criteria for your system’s rubric.
⚡ Latecomer shortcut — the minimum system of 3 boxes
Before building the perfect system, build the minimum system. 3 boxes. Ready to run now. In chat. It’s the MVP (minimum viable product) for a process — and it already delivers real value.
🧩 Workflow outline — Kit T5
Inject the baseline briefing. Provide the input material. Define what you want as the output for this stage.
[BOX 2 — AI generates a proposal]
AI produces the result based on the input. It doesn’t evaluate it yet.
[BOX 3 — Evaluation: criteria X + Y + Z]
→ ok: deliver the final result
→ not ok: make a surgical adjustment → go back to Box 2
📊 Why 3 boxes are enough to get started
- You can do it today. No approval, setup, or investment required.
- The minimum system shows you what to improve. After running it 3 times, you'll know which boxes need more detail.
- T6 (Evolution) takes care of the rest. You don’t need to get the system right in the first version—you just need it to exist.
Box 1 focuses on
Complete context. The richer it is, the better Box 2's proposal.
Box 2 focuses on
Generate without evaluating. Separating generation from evaluation improves both.
Box 3 focuses on
Objective criteria. This box is the system’s "quality control."
👥 In your profession — 5 workflows ready to adapt
The 3-box mini-system works in any field. Here are 5 workflows applied to 5 profiles — choose the closest match to yours and adapt it.
📚 Teacher — lesson creation workflow
C2: AI generates a lesson plan with objectives, activities, and assessment
C3: Measures: Aligned with the curriculum? ≤ 90 min? Includes a practical exercise? → ok/adjust
⚖️ Lawyer — contract review workflow
C2: AI identifies risky clauses, gaps, and negotiation suggestions
C3: Measures: Covered the 3 priority risks? Appropriate technical language? → ok/go deeper
💼 Salesperson — lead → proposal workflow
C2: AI generates a personalized proposal with a solution + benefits + CTA
C3: Measures: Does it address the stated pain point? Clear CTA? Right tone for the audience? → ok/adjust
📊 Manager — weekly decision workflow
C2: AI suggests 3 priorities for the week with justification
C3: Measures: Aligned with the goals? Within the constraints? Actionable? → ok/adjust
✏️ Creator — idea → published post workflow
C2: AI generates a hook + body + CTA in the network’s format
C3: Measures: Does the hook stop the scroll? Clear CTA? On-brand tone? ≤ 150 words? → ok/adjust
⚡ Slow User Shortcut — final version
Choose ONE of the workflows above. Adapt Box 1 to your real context. Run it today in a chat session. The first system that works is worth more than the perfect system that never got off the ground. T6 will teach you how to make it evolve with every cycle.
⚙️ Module Summary
📦 Module deliverable:
Run the 3-box mini-system from start to finish using a real case. Document: the 3 prompts you used, the quality criteria you defined, what passed, and what needed adjustment in the loop. This document is your documented mini-system — the T5 deliverable and the input foundation for T6.
Next Track:
T6 — Evolution Engineering: now that the system exists, how does it learn from each cycle and improve over time?