Amazon Bedrock

• 1. Introduction

  ›

  Amazon Bedrock was announced on the 13th of April 2023, a service from Amazon Web Services aimed at people who want to build generative artificial intelligence applications. The name sounds like geology, yet the product is pure cloud computing. What does it actually hand to a developer? Why would a company choose it over rival platforms? And what happens when the same platform also tries to let people with no programming background build software? The answers begin with a single promise about how many different models a customer can reach.

  12

• 2. One Interface, Many Models

  ›

  Amazon Web Services, usually shortened to AWS, built Bedrock around a unified API. Through that one interface, a developer reaches foundation models, often shortened to FMs, made by several different AI companies. Related tools sit alongside those models.

  Bedrock is also serverless. Customers do not have to set up or manage the machines that run their workloads. That makes it a cloud service in the fullest sense.

  The market is crowded. Bedrock competes with similar enterprise AI platforms, including Microsoft Foundry and Google Cloud Platform. Each rival offers its own way of packaging models for business customers, and Bedrock's pitch rests on choice among providers.

  123

• 3. From Announcement to Agents

  ›

  The 28th of September 2023 is the date Bedrock became generally available, roughly five months after the announcement. Until then, the service had been a preview for a limited audience. From that day, any AWS customer could build with it.

  The service kept growing. Across 2024 and 2025, AWS added AI agents to the platform. An agent is a model that can interact with external systems rather than only produce text.

  That shift matters because it moves a model from answering questions to acting on them. The next two features, though, deal with a quieter problem: what a model knows and what it is allowed to say.

  234

• 4. Private Data and Safety Controls

  ›

  Knowledge Bases is a feature built on Retrieval-Augmented Generation, known as RAG. It is a managed workflow, so customers do not have to assemble the pieces themselves. With it, a model can pull facts from private data stored in Amazon S3.

  Guardrails addresses safety and compliance. Administrators can set content filters, and they can have personally identifiable information redacted. The rules apply across all models on the platform, not one model at a time.

  Together, the two features answer the questions a cautious company asks first. Can the model use our own documents, and can we control its output? Bedrock's next product, aimed at a very different audience, asked neither.

  56

• 5. PartyRock

  ›

  In November 2023, Amazon launched PartyRock, a web-based no-code environment for building generative AI applications. Nobody needs to write code to use it. Instead, a person describes what they want in plain language, and the platform translates that description into software widgets.

  Each widget gives the application a specific behavior. Some run text-based prompts, and others act as conversational agents. Some generate images, while others summarize or answer questions about documents the user has uploaded.

  The pricing changed as the product matured. PartyRock began with a limited-time free trial. In early 2025, AWS moved it to a recurring free daily usage credit, so a user who returns tomorrow has fresh credit waiting.

  7