1.3.1. The AWS AI Stack: Bedrock, SageMaker AI, and Amazon Quick
💡 First Principle: Each layer of the AWS AI stack answers a different business question. "Do we want a ready-made tool our employees can use now?" points to applications such as Amazon Quick. "Do we want to build our own generative AI applications on proven models without managing infrastructure?" points to Amazon Bedrock. "Is the model itself our differentiator, built from our own data?" points to Amazon SageMaker AI.
Amazon Quick is the application layer named on the exam. It evolved from Amazon QuickSight into an AI-powered workspace for business users. It includes Quick Sight (interactive dashboards and business intelligence), Quick Research (in-depth research across the web and company data, delivered as cited reports), Quick Flows and Quick Automate (AI-driven workflows and process automation that take actions in connected applications), and Quick Index (connecting company documents so answers are grounded in internal information). It is licensed per user, which makes it the exam's typical example of seat-based pricing. Choose it when the goal is to put AI-assisted analysis and automation in employees' hands quickly, without a build project.
Amazon Bedrock is the managed platform for generative AI. It provides API access to a choice of foundation models from Amazon and other providers, so teams can build applications without hosting models. At a business level, know its building blocks: Knowledge Bases connect models to company content using Retrieval Augmented Generation (RAG); Guardrails filter harmful content, redact sensitive information, and check whether answers are grounded in source material; Agents and AgentCore let teams build and run AI agents that use tools and take actions. Bedrock is mainly consumption-based: you pay per token or request, with cheaper batch processing and an option to reserve capacity (Provisioned Throughput) for steady, high-volume workloads. Choose it when you want to build your own generative AI experience on proven models.
Amazon SageMaker AI is the custom ML layer: tools to build, train, deploy, and monitor your own models, including traditional ML such as demand forecasting, risk scoring, and recommendation. It is typically instance-based: you pay for the compute instances used while training and while models are hosted. Choose it when the model is the differentiator, when no managed model fits, or when your proprietary data is the source of advantage. It also includes SageMaker Clarify, which helps detect bias in data and models.
AWS Marketplace sits alongside all three: a catalog where organizations can buy third-party models, SaaS AI products, and datasets through their AWS account, which simplifies the buy side of build-buy-partner decisions.
⚠️ Exam Trap: "Generative AI" in a scenario does not automatically mean SageMaker AI, and "custom" does not automatically mean SageMaker AI either. Adding company documents to a generative AI assistant is a Bedrock Knowledge Bases (RAG) scenario. Training a new fraud-scoring model on years of proprietary transactions is a SageMaker AI scenario. Giving analysts AI-assisted dashboards this quarter is an Amazon Quick scenario.
Reflection Question: A regional insurer wants (a) underwriters to ask questions of policy documents, (b) a proprietary risk model trained on 20 years of claims, and (c) executives to get AI-generated insights from existing dashboards. Which layer fits each need, and how would each be priced?
