AWS connects Step Functions workflows to AI agents through AgentCore Gateway

AWS announced on October 29, 2026, that AI agents can use existing Step Functions Standard and Express workflows as tools through Amazon Bedrock AgentCore Gateway. The gateway connects agents to tools through the Model Context Protocol (MCP), according to the AWS announcement.

Agent decisions, defined workflows

The integration separates two responsibilities: the agent decides when to run a workflow and what input to supply; Step Functions executes the defined workflow. AWS also says agents can look up workflow runs, helping answer operational questions about which executions failed and where.

In its launch example, AWS describes a customer-support agent starting a refund workflow containing an eligibility check, manager approval and credit steps. The useful engineering boundary is between choosing an action and implementing the business process. Rather than asking a model to reconstruct that process on each request, teams can retain an explicit workflow and let the agent decide when to invoke it.

That separation should not be treated as a guarantee that every invocation is appropriate. A predefined sequence alone does not establish that the agent supplied valid inputs or had a sound reason to initiate a refund. Teams evaluating the integration should examine input validation, invocation permissions and approval requirements—not just workflow ordering.

Regional availability and service charges

AWS says the capability is available in all AWS Regions where Amazon Bedrock AgentCore Gateway is available, with standard Step Functions and AgentCore Gateway rates applying. The launch notice directs developers to a Step Functions tutorial and connector documentation for supported operations and setup options.

A practical starting point is a bounded workflow with clearly defined inputs and reviewable side effects. The central design question is which decisions belong to the agent—and which should remain encoded in the workflow.