AWS offers dozens of AI and machine-learning services, and the names don’t always make it obvious which to use. The good news is that they fall into three clear layers. Once you know the layers, choosing becomes much easier.
Layer 1: Ready-made AI applications
These are finished products for end users, needing little or no development.
- Amazon Q Business — a generative-AI assistant that answers questions and completes tasks using your company’s content and systems.
- Amazon Q Developer — an AI assistant for software development and AWS operations.
Choose this layer when a standard assistant meets the need and speed matters more than customisation.
Layer 2: Generative AI building blocks — Amazon Bedrock
Amazon Bedrock gives developers managed access to foundation models from several providers — including Anthropic Claude, Meta Llama, Mistral and Amazon Nova — through one API, together with Knowledge Bases for RAG, Agents and AgentCore for agentic applications, Guardrails for safety and tools for evaluation. Choose this layer when you are building your own generative-AI application or agent.
Layer 3: Machine-learning platform — Amazon SageMaker
SageMaker is the platform for data scientists and ML engineers to build, train, tune, deploy and monitor their own models, from classic forecasting and classification to fine-tuning open models. Choose this layer when you need custom models trained on your data, or full control over the ML lifecycle.
Purpose-built AI services
Alongside the three layers, AWS offers pre-trained services for specific tasks — often the fastest route for well-defined problems:
| Service | What it does | Typical use |
|---|---|---|
| Amazon Textract | Extracts text, forms and tables from documents | Invoices, claims, applications |
| Amazon Comprehend | Natural-language processing | Entity extraction, sentiment, classification |
| Amazon Transcribe | Speech to text | Call recordings, meetings |
| Amazon Polly | Text to speech | Voice applications, accessibility |
| Amazon Translate | Machine translation | Multilingual content and support |
| Amazon Rekognition | Image and video analysis | Content moderation, visual inspection |
| Amazon Personalize | Recommendations | Product and content recommendations |
How to choose
- Is there a ready-made assistant that fits? Start with Amazon Q.
- Is it a well-defined task like OCR or transcription? Use the purpose-built service.
- Do you need a custom generative-AI app or agent? Build on Bedrock.
- Do you need your own trained models? Use SageMaker.
Many real solutions combine layers — for example Textract to read documents, Bedrock to reason over them and Step Functions to orchestrate the workflow.
Our AWS and Amazon Bedrock teams help you pick and combine the right services. Service names and features evolve, so check current AWS documentation.
Frequently asked questions
What is the difference between Amazon Bedrock and SageMaker?
Bedrock provides managed access to pre-trained foundation models plus tools for RAG, agents and guardrails. SageMaker is a platform for building, training and deploying your own machine-learning models.
What is Amazon Q?
Amazon Q is a family of generative-AI assistants from AWS, including Amazon Q Business for answering questions over company content and Amazon Q Developer for software development.
Which AWS service should I use for document processing?
Amazon Textract extracts text, forms and tables; combining it with a Bedrock model adds classification, summarisation and reasoning over the extracted content.