AWS re:Invent 2025 - Kiro and Amazon Bedrock: Unlock AI Agents for Your Legacy Apps (MAM403)

Source: Dev.to

Overview

AWS re:Invent 2025 – Kiro and Amazon Bedrock: Unlock AI Agents for Your Legacy Apps (MAM403)

In this session, Ryan Peterson, AWS’s worldwide tech leader for modernization, explores the evolution from traditional application modernization to agentic AI systems. He traces AI development from the 2017 introduction of Transformers through tool use and the Model Context Protocol (MCP), predicting billions of agents by 2028. The demo showcases Amazon Bedrock AgentCore and AgentCore Gateway, which bridge AI agents to legacy applications using the MCP standard. Peterson also demonstrates how the open‑source Strands Agents SDK and Kiro IDE enable rapid agent development via spec‑based approaches. A live demo integrates an unmodified Swagger Petstore application with an AI agent, illustrating autonomous API queries, reasoning over results, and the combination of general knowledge with real‑time data—without changing existing code.

From Lambda to Agentic AI: A Decade of Application Modernization

Just before this session I was chatting with attendees who were experiencing their first re:Invent. It reminded me of my own first re:Invent in 2014, when Andy Jassy announced AWS Lambda followed shortly by Amazon ECS—a pivotal moment for application modernization.

For the next decade I worked with customers on cloud‑native architecture, serverless, containers, microservices, and distributed systems. Today, the next ten years will be defined by agentic AI.

I’ll begin with a primer on AI agents (first half) and then move to a live code walkthrough (second half). The goal is to explain why 2025 is the turning point for agentic AI systems across organizations.

The Evolution of AI: From Transformers to Autonomous Agents

Transformers (2017)

Transformers changed text processing by enabling parallel execution, dramatically increasing scalability compared with bag‑of‑words or n‑gram approaches.

Scaling Laws

As compute, parameters, and data grew, model performance improved predictably, encouraging heavy investment and further scaling.

Few‑Shot Reasoning & Chain‑of‑Thought Prompting

Larger models began to exhibit emergent reasoning abilities. By providing chain‑of‑thought prompts—explicitly describing reasoning steps—models dramatically improved at tasks like math without additional fine‑tuning.

Tool Use & the Rise of Agents

Enabling models to interact with external tools marked the birth of practical agents. The ReAct framework (Reason → Act → Interpret → React) formalized this loop, allowing agents to fetch real‑time data and act on it.

Model Context Protocol (MCP)

Released by Anthropic in late 2023, MCP provides a standard for passing context between models and tools, accelerating agent development and interoperability.