Stop Chatting with AI. Start Loops (Ralph Driven Development)
Replace interactive AI chat with structured execution loops—invest heavily in planning, dump full context each iteration, and let verification backpressure catch errors before they compound.
Core Insight
Typing creates friction that filters out context. Lost context causes hallucinations. The fix: speak your plans into existence, then run stateless loops where agents re-read everything each iteration.
The 5-Phase Methodology
1. Planning Phase
Invest substantial time (1+ hours) building a detailed plan.md before writing code. Use voice-to-text (handy.computer) to capture thoughts without the friction of typing.
2. Ralph Driven Development Bad AI results stem from bad prompts, bad context, or bad data access. Fix the prompt, not just the code. Named after Geoffrey Huntley's framework.
3. Execution Loop Agents re-read full context each iteration, preventing drift. Single task per iteration. Shell loops (Bash/PowerShell) automate the cycle.
4. Verification Backpressure Strict compilation rules and CLI verification reject errors before commits. This backpressure keeps the codebase viable.
5. Brownfield Application Human-in-the-loop review preserves viability for legacy codebases. Not everything can be fully automated.
Practical Setup
- handy.computer: Offline speech-to-text for rapid planning
- opencode.ai: Terminal agent runner
- AGENTS.md: Store operational insights for agents to reference
- CLI tools over MCP: Simpler, more portable
Connections
Builds on agentic-design-patterns and prompt chaining patterns from what-is-prompt-chaining-in-ai-agents. The backpressure concept aligns with context-efficient-backpressure.