AI Agents on AWS: Beginner's guide to building AI agents on AWS

“There are plenty of tutorials out there showing how to build a simple AI chatbot in a Jupyter notebook. But what happens when you actually need to deploy that system to enterprise production? That is exactly the gap AI Agents on AWS by Bunny Kaushik and Mona M attempts to close, and for the most part, it succeeds brilliantly.

The book's structure is incredibly well-paced. It takes you logically from the basics of Retrieval-Augmented Generation (RAG) right through tool-use, agent memory , and advanced orchestration patterns like Swarms and Directed Acyclic Graphs (DAGs). But the real standout sections are Chapters 6 and 7 , which focus on production deployment and AI governance. The authors do not shy away from the messy, unglamorous realities of enterprise AI. They tackle cold starts , the architectural trade-offs of stateless functions (Lambda) versus always-on containers (ECS) , and deterministic governance using Cedar policies. It is a highly production-first mindset that is often missing from AI literature.

I also really appreciated the protocol-agnostic thinking. The deep dive into the Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocol elevates this from a standard AWS how-to guide into a genuinely forward-looking architectural reference.

Do not let the "AWS" in the title fool you. The principles taught here are highly relevant outside the AWS ecosystem. The compute mapping transitions, such as moving from Lambda to ECS Fargate, perfectly mirror the choices you would make on GCP between Cloud Functions and Cloud Run or GKE. Because the authors emphasize OpenTelemetry (OTel), the instrumentation strategies are completely vendor-neutral. Traces sent to CloudWatch can easily be routed to Google Cloud Trace or Datadog , and the governance concepts behind Bedrock Guardrails map beautifully to GCP's Vertex AI Data Controls.

One opportunity for expansion is the book's reliance on the Strands SDK to simplify more intricate concepts. While this abstraction is excellent for teaching and accelerates the prototyping phase, it can sometimes obscure the finer details of state management and concurrency that enterprise architects ultimately need to master. Exploring the limitations of native frameworks, such as building with pure LangGraph or AutoGen without the Strands wrapper, would add a wonderful layer of technical depth.

Also, while Chapter 3 provides a solid primer on agent memory, it glosses over the headaches of state conflict resolution. What happens when multiple agents in a swarm concurrently try to read or write to the same long-term memory vector store or DynamoDB table? That is a massive real-world challenge that deserved more page space.

If you are a developer, cloud architect, or AI engineer looking to move your autonomous systems out of the sandbox and into scalable, secure, and observable production environments, this is an essential read.”

Khushan Adatiya, Senior Software Engineer, Google

What I appreciated most about AI Agents on AWS is that it meets you where you are. The language is accessible without being overbearing and complex concepts like multi-agent orchestration, the Model Context Protocol are explained with a clarity that makes them immediately actionable. The book earns its real value in the second half: the honest breakdown of when multi-agent architectures are the wrong choice, and the deterministic governance framework for controlling what agents can/must actually do (not just what they say), are the sections I'd recommend this book for. If you're an AWS practitioner looking for a single resource that connects the dots from agent fundamentals to production guardrails, this fills that gap well. Experienced builders may find the early chapters familiar territory, but gets you some payoff in the later sections like Chapter 7.

Gaurav Savla, AI Product Management Leader

Bunny Kaushik is a Solutions Architect at AWS and a AI/ML specialist who works with world's leading Financial Institutions to build scalable AI systems that deliver measurable business impact. He focuses on AI strategy, governance, and enterprise adoption, helping organizations transition from isolated prototypes to production-ready systems. He is also a published author, speaker, and active mentor to startups and professionals. He specializes in practical AI implementation, helping the next generation of builders navigate the transition from theory to production.

Mona is a Senior AI/ML Specialist Solutions Architect at AWS with over 15 years of experience spanning AI/ML, cloud, and software engineering. She specializes in designing and scaling production-grade AI systems, including LLMs, fine-tuning, inference optimization, and enterprise AI architecture across AWS and Google Cloud. Mona is the author of two bestselling books and is also a published speaker and researcher who has contributed to leading industry conferences and academic work. She actively mentors professionals in AI and cloud.