The Flywheel Planning Playbook¶

A step-by-step operational guide for planning and executing software projects with coding agents.

Based on Jeffrey Emanuel's (@doodlestein) agentic coding methodology and the Agentic Coding Flywheel. Prompt library derived from jeffreysprompts.com. Playbook derived from @voidserf's guide World-Class Planning With Agent Flywheel.

The Core Idea¶

Spend planning tokens to save implementation tokens. Reach a steady-state plan where remaining improvements are incremental, then unleash agents.

Planning is the highest-leverage activity in agent-driven development. A well-refined plan with proper task decomposition lets a swarm of agents execute in parallel without stepping on each other, missing requirements, or building the wrong thing.

The Workflow¶

graph TD
    P0[Phase 0: Prerequisites] --> P1[Phase 1: Draft the Plan]
    P1 --> P2[Phase 2: Refine the Plan]
    P2 -->|Loop until convergence| P2
    P2 --> P3[Phase 3: Convert Plan to Beads]
    P3 --> P4[Phase 4: QA the Beads]
    P4 -->|Loop until convergence| P4
    P4 --> P5[Phase 5: Implement with Agent Swarm]
    P5 --> P6[Phase 6: Fresh-Eyes Review]
    P6 -->|Loop until clean| P6

Each phase has a goal, steps, copy-paste prompts, and a stop condition that tells you when to move on.

Do not skip phases

Do not start implementation before your plan and beads stabilize. The whole point is front-loading the thinking.

What Makes This Different¶

Most developers jump straight to code. This playbook inverts the ratio:

Activity	Typical	Flywheel
Planning	~10% of effort	~85% of effort
Implementation	~70% of effort	~10% of effort
Review/Fix	~20% of effort	~5% of effort

The result: agents produce better code faster because they're working from a plan that has already been stress-tested through multiple critique rounds.

Getting Started¶

Read the Principles to understand the "why"
Walk through each phase starting with Phase 0
Use the Prompt Pack for ready-to-paste templates
Keep the Quick Reference card handy