The deliberate practice of maintaining human comprehension and control as AI capabilities accelerate, ensuring programmers remain creators rather than curators. ## Core Principle **Never commit code without full comprehension.** This is not about slowing down or rejecting AI assistance—it's about maintaining the boundary between augmentation and abdication. Human agency requires continued understanding, not delegation of judgment. ## The Practice Salvatore Zappalà models this standard through concrete behaviors: 1. **Maintain git control** - You decide what enters the codebase 2. **Refuse to commit without comprehension** - If you can't explain it, don't ship it 3. **Keep AI as collaborator, not driver** - Direction flows from human to AI, not reverse ## Why This Matters As LLMs evolve from code generation to autonomous planning (Claude Code's "Plan mode"), the velocity gap widens between what AI can produce and what humans can comprehend. Evidence: - Boris Cherny merged 259 PRs (497 commits, 40k lines) entirely written by Claude - "Vibe coding" creates compounding mistakes that "go down rabbit holes beyond recovery" - Runaway complexity exceeds even the model's holistic reasoning capacity The risk is not unemployment—it's becoming a "curator" who approves code without understanding it. ## The Philosophical Core > "I don't like the idea of a world where agency, in its true meaning, doesn't belong to humans anymore." > — Salvatore Zappalà This isn't anti-AI sentiment. It's recognition that authentic human agency requires the capacity to understand, evaluate, and direct—not just approve or reject. ## Cross-Domain Applications - **Knowledge Work**: Same principle applies to AI-generated analysis, writing, and decisions - **Strategic Planning**: Keep humans as architects, AI as implementation accelerant - **Skill Development**: Manual practice preserves capability; pure delegation erodes it (see [[DHH Warning on AI Usage]]) ## Related Concepts - [[Vibe Coding vs Mastery]] - The distinction this standard enforces - [[Amplified Intelligence]] - Human skill as the multiplier in AI output quality - [[AI/The Paradox of AI Agency - Augmentation vs Erosion]] - The broader philosophical tension - [[Human-AGI Collaboration Paradigm]] - Organizational framework for preserving human oversight ## Source [[Agency must stay with humans]] - Salvatore Zappalà, salvozappa.com, January 2025 --- *Atomic concept extracted January 2026*