Implement Two-Step Agentic Flow with Planning Phase

Overview

Enhance Canvas AI to use a two-step agentic workflow instead of the current single-step approach, introducing a planning phase before execution to improve AI decision making and task coordination.

Proposed resolution

Implement a two-phase agentic flow system:

Phase 1: Planning Agent

• Purpose: Analyze user requests and create detailed execution plans
• Responsibilities:
  • Break down complex requests into discrete tasks
  • Assess component dependencies and conflicts
  • Create execution roadmap
  • Validate feasibility before execution

Phase 2: Execution Agent

• Purpose: Execute the plan created by the planning agent
• Responsibilities:
  • Follow the detailed execution plan

User interface changes

• Enhanced progress indicators showing planning and execution phases using Polling (separate issue)

API changes

• New AI agent configuration for planning workflows
• Modified agent orchestration to support two-phase execution
• Enhanced function call sequencing and dependency management

Implementation details

TBC

Benefits

1. Better Planning: Complex tasks are better analyzed before execution
2. Reduced Errors: Planning phase catches issues before execution

Implementation notes

• Maintain backward compatibility with single-step workflows
• Consider plan caching for similar requests
• Implement proper timeout handling for both phases
• Ensure planning overhead doesn't significantly impact performance
• Add configuration options to enable/disable planning phase

Dependencies

• Requires ai_agents module
• Requires canvas_ai module
• May require enhanced temp store for plan persistence