AI-Assisted Development and Managed AI Services

What it is

This page covers two different ideas that share the word AI.

AI-assisted development tools help you write code. They suggest, generate, review and refactor code inside your editor, but they are not part of your application.

Managed AI services are APIs your application calls at run time. Amazon Bedrock generates text from foundation models. Amazon Textract reads documents. Amazon Transcribe turns speech into text. Amazon Comprehend analyzes text. You call them with an SDK, and AWS runs the models.

How it works

Tools that help you write code

Kiro is an agentic development environment from AWS, available as an IDE and a command line tool. Its signature feature is spec driven development. A spec turns an idea into three files: requirements.md (user stories and acceptance criteria), design.md (the technical design) and tasks.md (a list of trackable implementation tasks).

Kiro also reads steering files, markdown in .kiro/steering/ that record project conventions so you do not repeat them in every chat. Hooks are JSON definitions in .kiro/hooks/ that run an action when an event happens, such as a file save.

Amazon Q Developer provides inline code suggestions, chat, code generation and security scanning in supported IDEs and in the AWS Management Console. AWS has announced that the IDE plugins reach end of support on April 30, 2027, and points users to Kiro.

Treat generated code as a draft from a fast junior colleague. Read it, run the tests, run a security scan, and keep secrets and customer data out of prompts. A human still owns the pull request.

Calling Amazon Bedrock

Use the bedrock-runtime client. The Converse API sends messages in one format that works across models that support messages. InvokeModel sends a model specific JSON body instead. ConverseStream and InvokeModelWithResponseStream return the answer in chunks.

import boto3
br = boto3.client("bedrock-runtime")
resp = br.converse(
    modelId=MODEL_ID,
    messages=[{"role": "user", "content": [{"text": "Summarize this ticket: ..."}]}],
    inferenceConfig={"maxTokens": 300, "temperature": 0.2},
)
answer = resp["output"]["message"]["content"][0]["text"]

• The caller's role needs bedrock:InvokeModel (and bedrock:InvokeModelWithResponseStream for streaming) on the model.
• A guardrail that your team already configured is applied by passing its ID and version in the request, or by calling ApplyGuardrail.
• Throttling errors are retried with backoff like any SDK call.

Amazon Bedrock AgentCore is the platform for running AI agents. Its Runtime hosts them, Memory keeps context, Gateway turns APIs and Lambda functions into tools, and Observability traces them.

Document, speech and text services

Asynchronous jobs return a job ID. You learn that a job finished from an Amazon SNS message (Textract) or an EventBridge job state change event (Transcribe), or by polling the job status.

How the exam asks about this

• "A team wants to turn a feature idea into requirements, a design and tracked tasks before coding." -> Kiro spec driven development.
• "Extract tables and form fields from scanned invoices, with the least custom code." -> Amazon Textract.
• "Add summarization to an app and keep the same code if the model changes." -> Call the Bedrock Converse API.