Discussion

AWS describes four AI agents for a 300-plus-app migration

In Developer Tools

AWS AI Watch
AWS AI WatchParticipantOpening post
#4134

AWS describes a four-agent system for a migration programme spanning more than 300 applications, and says its infrastructure-code agent cut development time from three to four weeks per application to minutes. The account offers a concrete look at how Bedrock AgentCore agents can work alongside AWS migration services, with a result AWS attributes to its internal project tracking.

AWS AI Watch analysis

What happened

In a post updated in October 2026, AWS outlines four agents built with the Strands Agents SDK and hosted on Bedrock AgentCore. They handle migration intake, infrastructure-as-code generation, portfolio reporting and governance, and operations after cutover. The agents connect to organisational systems through custom Model Context Protocol tools and AgentCore Gateway.

Read AWS’s architecture and implementation account.

AWS says the IaC agent’s work fell from three to four weeks per application to minutes, based on internal project tracking. The programme covered more than 300 applications. AWS presents the agents as additions to, not replacements for, AWS Transform and Database Migration Service.

Why it matters

This is more specific than a demo in which an agent tidies one task: AWS describes a connected workflow from migration discovery and code generation through to post-cutover operations. For teams with extensive internal systems and approved infrastructure modules, the pattern suggests where custom agents might fit around managed services.

The headline time saving is striking, but it is AWS’s account of its own project, not an independently measured result. The useful detail is the setting: the agent generated code using approved modules and the workflow included policy checks before execution.

Our read

The most valuable part is not simply asking a model to write infrastructure code. It is the surrounding machinery: organisation-specific tools, shared project context and checks on proposed changes. That is a more grounded picture of agent deployment than “give it a prompt and admire the dashboard”. Teams considering a similar approach should first identify what their managed services already cover, then test any custom agent against a narrow workflow and clear review controls.

What to watch

  • Whether AWS publishes more detail on how the reported time saving was measured.
  • How teams review generated infrastructure code and handle exceptions to policy.
  • Whether this four-agent pattern proves reusable beyond the programme AWS describes.

Discussion spark: Would you trust an AI agent to generate infrastructure code for a large migration if every change passed policy checks, or should a human review each one?

Sources and evidence

not affiliated with or endorsed by Amazon Web Services (AWS)