Full worked example | Master AI Automation in 4 hours Master AI Automation in 4 hours Course About Ayush Modules Sample chapter Toolbox The Microcap Minute Classroom / Module 07: The AI Relay Team / Chapter 4 Full worked example Watch first, then read. Same lesson, your pace. What you will learn – The relay in action: draft → review → polish, with real handovers – How review legs actually change work – Timing and quota realities of a three-AI run The task Produce a one-page weekly tech newsletter for classmates: three stories, each with why-it-matters, plus a study tip. Budget: free tiers only. Team per our stack: Qwen drafts, Claude reviews, OpenAI polishes. Leg 1, Qwen drafts Fill the template; the essential parts: # HANDOVER, class-tech-newsletter · v1 ## GOAL A one-page weekly newsletter: 3 tech stories + 1 study tip. ## CONSTRAINTS Class-9 audience; ≤120 words per story; plain English; no hype words. ## YOUR TASK Draft all four sections. Pick stories from this week’s headlines you already know about; flag any fact you are not sure of with [CHECK]. ## OUTPUT CONTRACT Markdown only. Sections: Story1..3 (headline + body), StudyTip. Note [CHECK] : instructing drafters to self-flag uncertainty gives reviewers targets. Qwen returns v1, decent bones, maybe shaky facts. Leg 2, Claude reviews New baton: # HANDOVER, class-tech-newsletter · v2 (review leg) ## GOAL / CONSTRAINTS (unchanged, paste verbatim) ## ARTIFACT <<>> ## YOUR TASK Act as a strict editor. For each section verify factual plausibility, audience fit, word count. Produce: ## OUTPUT CONTRACT Verdict per section: PASS or ISSUES, then numbered issues tagged [fact] [tone] [length], each with a suggested fix. End with overall verdict: SHIP or FIX-FIRST. Claude returns something like: “Story2 [fact]: launch date conflicts with known timeline, suggest removing date entirely.” That single line justifies the whole relay, an error that would have shipped solo is now dead. Leg 3, OpenAI polishes Final baton carries everything: goal, constraints, the revised artifact with fixes applied , verboten (“don’t add new claims”), contract: “final clean markdown + 3-bullet changelog.” What comes back reads better than either predecessor, punchier headers, tighter close, with zero new facts smuggled in. File all three batons in projects/newsletter/ (h01-draft.md, h02-review.md, h03-final.md). Total human time: ~15 minutes, mostly pasting. Total AI time: minutes. Result: reviewed media on free tiers. Variations once fluent Swarm version: Kimi coordinates six parallel Qwen legs (one story each), merges, then Claude reviews the merged whole Two-AI version: any two models, draft+review, for smaller jobs Human-in-loop version: you play reviewer between legs, often the best use of your own attention Try it yourself Actually run this exact pipeline this week on a topic you care about. Time each leg. Then deliberately sabotage leg 2 by omitting the constraints section from its baton, observe the review quality drop. Restore it. File everything under projects/newsletter/ . You’ve now shipped the course’s signature workflow once; Module 12 will ask you to ship your own. Key takeaways – Draft → review → polish across three AIs produces reviewed output at ₹0. – Self-flagging instructions ([CHECK]) give reviewers handles. – Review legs earn their cost by killing errors before shipping. – Sabotage tests (drop constraints) teach more than success ever does. Download the exercise sheet (PDF) Module workbook (PDF) ← Prev: The Handover Doc Next: The review loop → Classroom / Module 07: The AI Relay Team / Chapter 4 Full worked example What you will learn – The relay in action: draft → review → polish, with real handovers – How review legs actually change work – Timing and quota realities of a three-AI run The task Produce a one-page weekly tech newsletter for classmates: three stories, each with why-it-matters, plus a study tip. Budget: free tiers only. Team per our stack: Qwen drafts, Cl
Full worked example
Written by
in