ai-dlc-fundamentals

# AI-DLC Fundamentals

AI-DLC (AI-Driven Development Lifecycle) is a methodology for collaborative human-AI software development. It addresses the fundamental challenge of maintaining productive AI sessions across context window limitations.

## Core Philosophy

### The Context Problem

AI coding assistants face a fundamental limitation: context windows are finite. As sessions grow longer:
- Context accumulates (code, errors, conversation history)
- Signal-to-noise ratio decreases
- AI may "forget" earlier decisions or repeat mistakes
- Quality of suggestions degrades

Traditional approaches try to work around this by:
- Larger context windows (expensive, diminishing returns)
- Better summarization (lossy, loses nuance)
- Retrieval augmentation (latency, relevance issues)

### The AI-DLC Solution

AI-DLC takes a different approach: **embrace context resets as a feature, not a bug**.

Instead of fighting context limits:
1. **Plan for iterations** - Work in deliberate cycles
2. **Preserve state externally** - Store intent, criteria, and learnings outside the context
3. **Fresh starts are good** - Each iteration begins with clean context + injected state
4. **Files are memory** - Persist what matters between sessions

## The Three Pillars

### 1. Backpressure Over Prescription

Traditional development processes prescribe steps:
- "Write tests first"
- "Get code review before merge"
- "Run linting before commit"

These become checkbox exercises that teams learn to game.

AI-DLC uses **backpressure** instead:
- Quality gates that **block progress** until satisfied
- Automated enforcement via hooks
- The AI learns to satisfy constraints, not follow scripts

Example backpressure:
```bash
# Stop hook that fails if tests don't pass
bun test || exit 1
```

The AI can't complete work until tests pass. It learns to write tests and fix failures, not because a process document says to, but because the system won't let it proceed otherwise.

### 2. Completion Criteria Enable Auton