Architecture Deep Dive

How AWS Transform
Works in Kiro

Two agents, one workflow. The Kiro IDE orchestrates locally while Bedrock AgentCore reasons in the cloud. Your code never leaves your machine.

01

Two Agents, One Workflow

AWS Transform splits intelligence (cloud) from execution (local). The Task Agent reasons about your code; the CLI applies changes on your machine.

YOUR MACHINE — LOCAL ZONE Kiro Agent Orchestrator & HITL Proxy Orchestrator ATX CLI Executor & Build Runner Executor Your Code Files never leave your machine Private API calls ↓ ↑ Results AWS CLOUD — BEDROCK AGENTCORE Bedrock AgentCore Foundation model reasoning Planning & code generation Task Agent Analyzes code, builds plan Guides each transform step Knowledge Items Learned patterns from every execution TD Defs Transform definitions
02

Three Execution Modes

Choose your level of control — from fully autonomous pipelines to interactive IDE-driven workflows.

Human "Upgrade to Java 21" Kiro Agent HITL loop ATX CLI AgentCore Kiro answers agent questions automatically — no human wait time
You approve plans asks questions ATX CLI AgentCore You approve every plan and answer every question — maximum control
CI/CD Pipeline No human needed ATX CLI AgentCore ×128 parallel repos on AWS Fargate No prompts — agent decides everything. Embed in CI/CD for enterprise scale.
02b

Any Tool. Any Agent. One Platform.

AWS Transform meets you where you work — invoke it from your IDE, CLI, web console, or any MCP-compatible tool. Extend it with your own agents.

Web Console

Purpose-built web experience for business stakeholders and technical teams. Manage large-scale campaigns, track progress, collaborate across workspaces.

Kiro IDE

Pre-built playbooks and conversational interface. Start in Kiro, check progress in the web console, finish in your IDE. Context is never lost.

MCP / A2A Gateway

Invoke AWS Transform from Claude Code, Cursor, Codex, GitHub Copilot, or your own orchestration framework — via Model Context Protocol or the AWS Transform agent skill. Agent-to-Agent communication built in.

NEW

ATX CLI

Scriptable, composable CLI for automation pipelines. Run interactively or autonomously. Embed into CI/CD, containers, or batch orchestration.

Visual Studio

IDE extension for .NET developers. Transform applications directly within your development environment without context-switching.

Agent Builder

Create custom agents with the Strands Agents framework and AgentCore SDK. Your agents plus AWS agents — one platform, fully extensible.

NEW

All data stays centralized. Regardless of which surface you use — web, IDE, CLI, or MCP — your transformation state, knowledge items, and progress are shared. Start a transformation in Kiro, monitor it in the web console, review results in your IDE. Your team always has visibility.

03

The Learning Loop

Every transformation makes the next one better. The agent captures debugging steps, human input, and code patterns as reusable knowledge.

Knowledge Items Define TD + rules Execute Transform code Verify Build & test Learn Save patterns
04

Self-Debugging in Action

When a build fails, the agent doesn't stop — it reads the error, reasons about the fix, applies it, and tries again. Watch the loop play out:

Agent edits code
Runs build
Build fails
Reads error
Reasons about fix
Applies fix
Runs build again
Build passes!
Watch the self-debugging loop play automatically...