AWS Services for Building GenAI Apps

AWS offers several services for building generative AI (GenAI) applications, and they sit at different levels. Some give you models, some help you build and run agents, and some are finished assistants that people just use. Exam questions usually describe what a team wants to build and how much it wants to manage, then ask which service fits. This lesson sorts the services by the job each one does.

Models Through an API with Amazon Bedrock

Amazon Bedrock is a fully managed service that gives you foundation models (FMs) from many providers through one API. There are no servers to run, and you can swap between models without rebuilding your application. Bedrock also includes the building blocks most GenAI apps need: Knowledge Bases for retrieval augmented generation (RAG), Guardrails for safety, model evaluation, and customisation such as fine-tuning.

Amazon Nova is Amazon's own family of foundation models, used through Bedrock. The current generation, Nova 2, has three models:

• Nova 2 Lite reads text, images, video and documents and answers with text, tuned for speed and low cost.
• Nova 2 Sonic holds real-time spoken conversations, taking speech in and replying with speech.
• Nova Multimodal Embeddings turns text, images, documents, video and audio into embeddings for search and recommendations.

The first-generation image and video models, Nova Canvas and Nova Reel, reached end of life on Bedrock in September 2026.

Building Your Own Models with SageMaker AI

Amazon SageMaker AI is the platform for building, training and deploying your own machine learning models, with full control over data, algorithms and infrastructure. SageMaker JumpStart, inside SageMaker AI, is a hub of pre-trained and open-source models, solution templates and example notebooks. You can deploy, fine-tune or evaluate a JumpStart model on endpoints in your own account.

The rule of thumb: Bedrock when you want to use models with the least effort, SageMaker AI when you need to build or closely control them.

Building and Running Agents

An agent uses a model to plan steps and call tools to complete a task. AWS splits agent work into writing the agent and running it:

• Strands Agents is an open-source software development kit (SDK) for writing agents. You define an agent as a model, a set of tools and a prompt, and the SDK runs the loop of reasoning and tool calls. It works with Bedrock models and other providers, and it can use Model Context Protocol (MCP) servers as tools.
• Amazon Bedrock AgentCore is a platform for deploying and operating agents securely at scale, with no infrastructure to manage. It works with any framework (Strands, LangGraph, CrewAI and others) and any model. Its modular services include Runtime (serverless hosting with isolated sessions), Memory (short- and long-term), Gateway (turns APIs and Lambda functions into MCP tools), Identity (agent authentication and access), Observability (tracing and monitoring), Policy (rules on which tools an agent may use), plus Browser and Code Interpreter tools.

Ready-Made AI Assistants

Some services are finished products rather than building blocks:

• Amazon Q Developer is an AI assistant for building and operating on AWS. It answers questions about AWS resources and architecture, and helps with code. AWS ends support for its IDE plugins on April 30, 2027 and points IDE users to Kiro.
• Amazon Quick is an AI assistant for business users: ask questions of company data, research, build dashboards and automate tasks.
• Kiro is an agentic coding service that turns requests into specifications, implementation steps, code and tests.

On the Exam

The common trap is mixing up Strands Agents and AgentCore. Strands is how you write the agent; AgentCore is where you run it.