Multi-Agent Systems

Why Multiple Agents?

Complex tasks often benefit from specialized agents working together, just like a team of experts. Each agent can have different tools, knowledge, and reasoning strategies.

Common Patterns

1. Supervisor Pattern

A "manager" agent delegates tasks to specialized worker agents:

Supervisor → Research Agent (web search, documents)
           → Code Agent (write and test code)
           → Review Agent (check quality, suggest improvements)

2. Debate/Consensus

Multiple agents independently solve a problem, then discuss and converge on the best answer. Improves accuracy on complex reasoning tasks.

3. Pipeline

Agents process information sequentially, each adding their expertise:

Analyst → Planner → Implementer → Reviewer → Deployer

4. Swarm

Agents hand off to each other based on the current need. Lightweight, flexible routing.

Frameworks

• Claude's tool_use: Native multi-turn tool use with parallel execution
• LangGraph: State machine-based agent orchestration
• CrewAI: Role-based multi-agent framework
• AutoGen: Microsoft's conversational multi-agent framework
• OpenAI Swarm: Lightweight agent hand-off pattern

Challenges

• Cost: Multiple LLM calls per task multiply API costs
• Latency: Sequential agent calls add up quickly
• Error propagation: One agent's mistake can cascade
• Debugging: Complex interactions are hard to trace
• Coordination: Agents may conflict or duplicate work

🌼 Daisy+ in Action: Multi-Agent Collaboration

Daisy+ implements multi-agent collaboration natively: DaisyBot handles customer-facing livechat, specialized digital employees manage domain-specific tasks (accounting, project management), and they communicate through Discuss channels — the same messaging system human employees use. An email can trigger one agent to create a task, another to draft a response, and a third to update the CRM — all coordinated through the ERP's own messaging infrastructure.