AI Responsibilities in DDAD
Overview
In DDAD, AI is a governed engineering participant that executes design, not an autonomous author that makes decisions.
Core Responsibilities
1. Follow AGENTS.md Constraints
What: Read and follow AGENTS.md governance rules
Activities:
- Read AGENTS.md before starting work
- Understand capabilities and limitations
- Follow constraints strictly
- Respect boundaries
Why: AGENTS.md governs AI behavior and prevents scope creep.
2. Execute Exactly One TODO Item at a Time
What: Work on exactly one authorized TODO item
Activities:
- Find TODO items with Status: TODO
- Check dependencies are DONE
- Select one TODO item
- Execute completely
- Don’t start next until current is DONE
Why: Focused execution ensures quality and prevents scope creep.
3. Implement Strictly According to LLD
What: Follow LLD Markdown exactly
Activities:
- Read referenced LLD section
- Understand requirements
- Implement exactly as specified
- Don’t add features not in LLD
- Don’t omit features from LLD
Why: LLD defines behavior. Following LLD ensures correct implementation.
4. Avoid Speculative or Unrelated Changes
What: Only make changes related to the TODO item
Activities:
- Stay within TODO scope
- Don’t add unrelated features
- Don’t make speculative improvements
- Don’t refactor unrelated code
Why: Scope control prevents architectural drift and maintains focus.
5. Commit Changes Transparently
What: Commit all changes with clear messages
Activities:
- Stage all changes
- Write clear commit messages
- Reference TODO item
- Commit transparently
Why: Transparent commits enable review and audit.
Additional Responsibilities
Run Tests or Validations
- Run unit tests
- Run integration tests
- Validate against LLD
- Check code quality
Update TODO Status
- Move TODO to DONE when complete
- Update TODO.md
- Maintain status
Generate Tests
- Create unit tests
- Create integration tests
- Ensure coverage
- Follow testing standards
What AI Doesn’t Do
❌ Make Design Decisions
- AI doesn’t decide what to build
- Humans define design in LLD
- AI only implements
❌ Authorize Work
- AI doesn’t create TODO items
- Humans authorize via TODO.md
- AI only executes authorized work
❌ Skip Review
- AI doesn’t bypass human review
- All code must be reviewed
- Humans approve changes
❌ Work Autonomously
- AI doesn’t work without authorization
- Always needs TODO item
- Always follows governance
The Mental Model
AI is a governed engineering participant,
not an autonomous author.
AI executes design. Humans define design.
Key Principles
1. Explicit Authorization Required
- Never assume authorization
- Always check TODO.md
- Only work on authorized items
2. Follow LLD Exactly
- Implement strictly according to LLD
- No additions outside LLD
- No omissions from LLD
3. One TODO at a Time
- Focus on one item
- Complete before moving on
- Maintain scope
4. Transparent Execution
- Commit all changes
- Clear commit messages
- Reference TODO and LLD
Best Practices
1. Read Artifacts Carefully
- Read AGENTS.md for constraints
- Read TODO.md for authorization
- Read LLD for requirements
2. Execute Completely
- Finish TODO completely
- Don’t leave partial work
- Ensure tests pass
3. Stay Within Scope
- Only work on authorized TODO
- Don’t add unrelated features
- Maintain focus
4. Commit Transparently
- Clear commit messages
- Reference TODO item
- Document changes
Common Mistakes
❌ Working Without Authorization
Problem: Starting work without TODO item
Solution: Always check TODO.md first
❌ Adding Features Not in LLD
Problem: Implementing features not specified
Solution: Follow LLD exactly, no additions
❌ Working on Multiple TODOs
Problem: Trying to do multiple items at once
Solution: One TODO at a time, complete before next
❌ Skipping Tests
Problem: Not running tests before commit
Solution: Always run tests, ensure they pass