workflow:
  id: llm-agent-greenfield
  name: Research-Driven LLM Agent Development
  description: Streamlined workflow for developing new AI agents with emphasis on research-driven decisions and adaptive development.
  type: greenfield
  project_types:
    - conversational-ai
    - task-automation
    - analytical-agent
    - creative-agent
    - multi-agent-system
    - voice-agent
  approach: Research-first development with adaptive phases based on project discovery and current best practices
  key_phases:
    discovery:
      description: Research requirements and assess feasibility
      agent: llm-architect
      actions:
        - Research current AI agent patterns relevant to use case
        - Analyze requirements and technical feasibility
        - Investigate existing solutions and architectural approaches
        - Define success criteria and project constraints
    design:
      description: Research-driven architecture and prompt design
      agents: [llm-architect, llm-engineer]
      actions:
        - Research architectural patterns and design solutions based on findings
        - Investigate prompt engineering approaches for the specific use case
        - Design safety and monitoring frameworks based on current standards
        - Create implementation plan with research-backed technology choices
    develop:
      description: Implement and test using research-backed methodologies
      agents: [llm-engineer, qa, llm-safety-governance]
      actions:
        - Research development frameworks and implement based on current best practices
        - Apply research-driven prompt engineering and optimization techniques
        - Implement comprehensive testing using current validation methodologies
        - Conduct safety reviews and validation based on current standards
    deploy:
      description: Deploy with monitoring using current deployment patterns
      agent: llm-engineer
      actions:
        - Research deployment strategies and implement production-ready solution
        - Set up monitoring and observability based on current best practices
        - Validate production performance and establish feedback loops
        - Document implementation and lessons learned