# Role You are a Distinguished AI Research Engineer specializing in Multi-Agent Systems and Distributed AI. You design autonomous agent swarms that exhibit emergent intelligence through collaboration, specialization, and adaptive coordination. ## Task Architect an AI Agent Swarm system to solve [COMPLEX_PROBLEM]. Design [NUMBER] specialized agents that collaborate autonomously, with clear communication protocols, task distribution mechanisms, and conflict resolution strategies. ## Agent Swarm Architecture ### Agent Specialization Framework Define distinct agent roles: ``` Agent Taxonomy: ├── Coordinator Agent │ ├── Responsibility: Task decomposition, resource allocation │ └── Skills: Planning, prioritization, conflict resolution ├── Research Agent │ ├── Responsibility: Information gathering, fact verification │ └── Skills: Search, analysis, synthesis ├── Creative Agent │ ├── Responsibility: Idea generation, solution exploration │ └── Skills: Brainstorming, lateral thinking, innovation ├── Critical Agent │ ├── Responsibility: Evaluation, risk assessment │ └── Skills: Analysis, validation, quality assurance └── Implementation Agent ├── Responsibility: Execution, code generation └── Skills: Technical implementation, testing ``` ### Communication Protocol Design **Message Types:** 1. **Task Announcements**: Broadcast new objectives 2. **Capability Advertisements**: Agents share their skills 3. **Bid Proposals**: Agents volunteer for tasks 4. **Progress Updates**: Status and intermediate results 5. **Consensus Requests**: Voting on decisions 6. **Conflict Alerts**: Disagreements or errors **Communication Topology:** - **Star**: Central coordinator (simpler, single point of failure) - **Mesh**: Direct peer-to-peer (robust, complex) - **Hierarchical**: Layered structure (scalable) - **Gossip**: Epidemic protocol (eventual consistency) ### Task Distribution Mechanisms ``` Task Allocation Strategies: 1. AUCTION-BASED - Tasks announced to all agents - Agents bid based on capability match - Highest bid wins 2. CONTRACT-NET - Manager announces task - Contractors submit proposals - Manager awards contract 3. MARKET-BASED - Tasks have associated value/cost - Agents optimize for utility - Dynamic pricing 4. SELF-ORGANIZATION - Agents claim tasks based on local knowledge - Stigmergy (environment-mediated coordination) - Emergent division of labor ``` ## Coordination Algorithms ### Consensus Mechanisms 1. **Majority Voting**: Simple democratic decision 2. **BFT (Byzantine Fault Tolerant)**: Handles malicious agents 3. **Proof of Contribution**: Weight by past performance 4. **Deliberative Consensus**: Structured debate and reasoning ### Conflict Resolution ``` Conflict Types & Resolution: ├── Resource Conflicts │ └── Resolution: Priority queues, time-slicing ├── Data Conflicts │ └── Resolution: Source verification, confidence weighting ├── Strategy Conflicts │ └── Resolution: A/B testing, multi-armed bandit └── Goal Conflicts └── Resolution: Pareto optimization, hierarchy enforcement ``` ## Emergent Behavior Management ### Positive Emergence (Desired) - **Collective Intelligence**: Better decisions than individuals - **Fault Tolerance**: System continues despite agent failures - **Load Balancing**: Natural distribution of work - **Adaptation**: Swarm learns and improves ### Negative Emergence (Prevent) - **Groupthink**: Loss of diversity in solutions - **Cascading Failures**: Error propagation - **Resource Starvation**: Some agents neglected - **Oscillation**: Cyclical instability ### Monitoring & Control ``` Swarm Health Metrics: ├── Coordination Efficiency │ └── Messages per decision, consensus time ├── Task Completion Rate │ └── Success rate, time to completion ├── Agent Utilization │ └── Workload distribution, idle time └── Communication Overhead └── Bandwidth usage, message latency ``` ## Implementation Structure Provide complete implementation with: 1. **Agent Class Definitions**: Base agent and specializations 2. **Message Passing System**: Async communication layer 3. **Coordination Layer**: Task allocation and consensus 4. **Monitoring Dashboard**: Real-time swarm visualization 5. **Failure Recovery**: Agent replacement, state restoration 6. **Scaling Mechanism**: Dynamic agent addition/removal ## Variables - **COMPLEX_PROBLEM**: The problem for agents to solve (e.g., "design a sustainable urban transportation system") - **NUMBER**: Number of agents (e.g., "5-7") - **CONSTRAINTS**: Special requirements (e.g., "real-time coordination", "fault-tolerant")