Skip to the content.

DDAD Execution Flow

Overview

The DDAD execution flow defines the step-by-step process of how humans and AI collaborate to build software using design artifacts.

The Flow

1. Human defines design in LLD Markdown
    ↓
2. Human adds a TODO item referencing the LLD
    ↓
3. AI reads AGENTS.md for constraints
    ↓
4. AI executes the TODO according to the LLD
    ↓
5. AI runs tests or validations
    ↓
6. AI commits changes
    ↓
7. AI moves the TODO item to DONE
    ↓
8. Human reviews and continues

Step-by-Step Details

Step 1: Human Defines Design in LLD Markdown

Who: Human
What: Create or update LLD Markdown document
Why: Define exact implementation requirements

Activities:

Output: LLD Markdown document

Step 2: Human Adds TODO Item

Who: Human
What: Add TODO item to TODO.md
Why: Authorize specific work

Activities:

Output: Updated TODO.md

Step 3: AI Reads AGENTS.md

Who: AI
What: Read AGENTS.md for constraints
Why: Understand governance and boundaries

Activities:

Output: AI understands what it can and cannot do

Step 4: AI Executes TODO According to LLD

Who: AI
What: Implement according to LLD
Why: Generate code that matches design

Activities:

Output: Generated code

Step 5: AI Runs Tests or Validations

Who: AI
What: Run tests and validations
Why: Ensure code works correctly

Activities:

Output: Test results

Step 6: AI Commits Changes

Who: AI
What: Commit changes to repository
Why: Record work transparently

Activities:

Output: Committed code

Step 7: AI Moves TODO to DONE

Who: AI
What: Update TODO status
Why: Track completion

Activities:

Output: Updated TODO.md

Step 8: Human Reviews Design and Verifies Code

Who: Human
What: Review design and verify code conformance
Why: Ensure design correctness and code matches design

Activities:

Output: Design reviewed and code verified (or design changes requested)

Note: In DDAD, design is reviewed and code is verified. No AI code review tools are used. Code correctness is determined by conformance to design.

Key Principles

1. Explicit Authorization

2. One TODO at a Time

3. Follow LLD Exactly

4. Transparent Execution

Example Flow

Scenario: Create User Registration Endpoint

  1. Human: Creates LLD.md section “User Registration Service”
  2. Human: Adds TODO item “Create User Registration Endpoint” referencing LLD
  3. AI: Reads AGENTS.md, understands it’s a Backend Service Agent
  4. AI: Reads LLD section, understands requirements
  5. AI: Generates endpoint code according to LLD
  6. AI: Runs tests, validates code
  7. AI: Commits code with message “Implement user registration endpoint (TODO: Create User Registration Endpoint)”
  8. AI: Updates TODO.md, moves item to DONE
  9. Human: Reviews code, compares to LLD, approves
  10. Human: Continues with next TODO item

Common Patterns

Pattern 1: Sequential Execution

TODO 1 → DONE → TODO 2 → DONE → TODO 3 → DONE

Execute TODO items one at a time in order.

Pattern 2: Parallel Execution

TODO 1 → DONE
TODO 2 → DONE  (can execute in parallel if no dependencies)
TODO 3 → DONE

Execute independent TODO items in parallel.

Pattern 3: Dependent Execution

TODO 1 → DONE → TODO 2 (depends on 1) → DONE → TODO 3 (depends on 2) → DONE

Execute TODO items respecting dependencies.

Best Practices

1. Complete Before Moving On

2. Review Regularly

3. Update Artifacts

4. Document Decisions