Summary

Upgrade the LangGraph agent backend from the current create_agent pattern to LangChain deep agents, which provide enhanced reasoning, multi-step planning, and tool-use capabilities.

Current State

• Agent is defined in apps/agent/main.py using langchain.agents.create_agent with ChatOpenAI
• Uses CopilotKitMiddleware for frontend state sync
• Tools: todo management, data queries, form generation, templates, visualization skills

Proposed Changes

• Upgrade langchain / langgraph packages to versions supporting deep agents
• Refactor apps/agent/main.py to use the deep agent API
• Verify CopilotKit middleware compatibility with the new agent architecture
• Update agent state schema if needed for deep agent state management
• Test bidirectional state sync (todos, templates, generative UI) still works correctly
• Update environment/dependency files (pyproject.toml, lock files)

Motivation

Deep agents offer improved reasoning and multi-step planning which would make the demo more compelling — especially for complex todo organization, multi-tool workflows, and richer generative UI responses.