کتاب ساخت برنامه‌ها با ایجنت‌های هوش مصنوعی

مقدمه‌ای بر ایجنت‌ها

Defining AI Agents • The Pretraining Revolution • Types of Agents • Model Selection • From Synchronous to Asynchronous Operations • Practical Applications and Use Cases • Workflows and Agents • Principles for Building Effective Agentic Systems • Organizing for Success in Building Agentic Systems • Agentic Frameworks • LangGraph • AutoGen • CrewAI • OpenAI Agents Software Development Kit (SDK) • Conclusion

طراحی سیستم‌های ایجنت

Our First Agent System • Core Components of Agent Systems • Model Selection • Tools • Designing Capabilities for Specific Tasks • Tool Integration and Modularity • Memory • Short-Term Memory • Long-Term Memory • Memory Management and Retrieval • Orchestration • Design Trade-Offs • Performance: Speed/Accuracy Trade-Offs • Scalability: Engineering Scalability for Agent Systems • Reliability: Ensuring Robust and Consistent Agent Behavior • Costs: Balancing Performance and Expense • Architecture Design Patterns • Single-Agent Architectures • Multiagent Architectures: Collaboration, Parallelism, and Coordination • Best Practices • Iterative Design • Evaluation Strategy • Real-World Testing • Conclusion

طراحی تجربه کاربری برای سیستم‌های ایجنت‌محور

Interaction Modalities • Text-Based • Graphical Interfaces • Speech and Voice Interfaces • Video-Based Interfaces • Combining Modalities for Seamless Experiences • The Autonomy Slider • Synchronous Versus Asynchronous Agent Experiences • Design Principles for Synchronous Experiences • Design Principles for Asynchronous Experiences • Finding the Balance Between Proactive and Intrusive Agent Behavior • Context Retention and Continuity • Maintaining State Across Interactions • Personalization and Adaptability • Communicating Agent Capabilities • Communicating Confidence and Uncertainty • Asking for Guidance and Input from Users • Failing Gracefully • Trust in Interaction Design • Conclusion

استفاده از ابزارها

LangChain Fundamentals • Local Tools • API-Based Tools • Plug-In Tools • Model Context Protocol • Stateful Tools • Automated Tool Development • Foundation Models as Tool Makers • Real-Time Code Generation • Tool Use Configuration • Conclusion

ارکستراسیون

Agent Types • Reflex Agents • ReAct Agents • Planner-Executor Agents • Query-Decomposition Agents • Reflection Agents • Deep Research Agents • Tool Selection • Standard Tool Selection • Semantic Tool Selection • Hierarchical Tool Selection • Tool Execution • Tool Topologies • Single Tool Execution • Parallel Tool Execution • Chains • Graphs • Context Engineering • Conclusion

دانش و حافظه

Foundational Approaches to Memory • Managing Context Windows • Traditional Full-Text Search • Semantic Memory and Vector Stores • Introduction to Semantic Search • Implementing Semantic Memory with Vector Stores • Retrieval-Augmented Generation • Semantic Experience Memory • GraphRAG • Using Knowledge Graphs • Building Knowledge Graphs • Promise and Peril of Dynamic Knowledge Graphs • Note-Taking • Conclusion

یادگیری در سیستم‌های ایجنت‌محور

Nonparametric Learning • Nonparametric Exemplar Learning • Reflexion • Experiential Learning • Parametric Learning: Fine-Tuning • Fine-Tuning Large Foundation Models • The Promise of Small Models • Supervised Fine-Tuning • Direct Preference Optimization • Reinforcement Learning with Verifiable Rewards • Conclusion

از یک ایجنت تا چندین ایجنت

How Many Agents Do I Need? • Single-Agent Scenarios • Multiagent Scenarios • Swarms • Principles for Adding Agents • Multiagent Coordination • Democratic Coordination • Manager Coordination • Hierarchical Coordination • Actor-Critic Approaches • Automated Design of Agent Systems • Communication Techniques • Local Versus Distributed Communication • Agent-to-Agent Protocol • Message Brokers and Event Buses • Actor Frameworks: Ray, Orleans, and Akka • Orchestration and Workflow Engines • Managing State and Persistence • Conclusion

اعتبارسنجی و اندازه‌گیری

Measuring Agentic Systems • Measurement Is the Keystone • Integrating Evaluation into the Development Lifecycle • Creating and Scaling Evaluation Sets • Component Evaluation • Evaluating Tools • Evaluating Planning • Evaluating Memory • Evaluating Learning • Holistic Evaluation • Performance in End-to-End Scenarios • Consistency • Coherence • Hallucination • Handling Unexpected Inputs • Preparing for Deployment • Conclusion

نظارت در محیط تولید

Monitoring Is How You Learn • Monitoring Stacks • Grafana with OpenTelemetry, Loki, and Tempo • ELK Stack (Elasticsearch, Logstash/Fluentd, Kibana) • Arize Phoenix • SigNoz • Langfuse • Choosing the Right Stack • OTel Instrumentation • Visualization and Alerting • Monitoring Patterns • Shadow Mode • Canary Deployments • Regression Trace Collection • Self-Healing Agents • User Feedback as an Observability Signal • Distribution Shifts • Metric Ownership and Cross-Functional Governance • Conclusion

حلقه‌های بهبود

Feedback Pipelines • Automated Issue Detection and Root Cause Analysis • Human-in-the-Loop Review • Prompt and Tool Refinement • Aggregating and Prioritizing Improvements • Experimentation • Shadow Deployments • A/B Testing • Bayesian Bandits • Continuous Learning • In-Context Learning • Offline Retraining • Conclusion

محافظت از سیستم‌های ایجنت‌محور

The Unique Risks of Agentic Systems • Emerging Threat Vectors • Securing Foundation Models • Defensive Techniques • Red Teaming • Threat Modeling with MAESTRO • Protecting Data in Agentic Systems • Data Privacy and Encryption • Data Provenance and Integrity • Handling Sensitive Data • Securing Agents • Safeguards • Protections from External Threats • Protections from Internal Failures • Conclusion

همکاری انسان و ایجنت

Roles and Autonomy • The Changing Role of Humans in Agent Systems • Aligning Stakeholders and Driving Adoption • Scaling Collaboration • Agent Scope and Organizational Roles • Shared Memory and Context Boundaries • Trust, Governance, and Compliance • The Lifecycle of Trust • Accountability Frameworks • Escalation Design and Oversight • Privacy and Regulatory Compliance • Conclusion: The Future of Human-Agent Teams