---
type: "post"
id: "aws-step-functions-workflows-can-now-be-used-33d8c0"
slug: "aws-step-functions-workflows-can-now-be-used-33d8c0"
title: "AWS Step Functions workflows can now be used as AI agent tools via Amazon Bedrock AgentCore Gateway"
description: "AWS has made existing AWS Step Functions workflows available as tools for AI agents through Amazon Bedrock AgentCore Gateway."
retrieval_nugget: "AWS has made existing AWS Step Functions workflows available as tools for AI agents through Amazon Bedrock AgentCore Gateway. The change matters because it lets an agent decide whether to initiate a business process while Step Functions executes that process in its predefined order, rather than leaving the model to sequence operational steps itself."
published_at: "2026-10-29"
updated_at: "2026-10-29"
record_date: "2026-10-29"
date_kind: "published_at"
topics: ["agent-tools","mcp","skills"]
entities: ["Amazon Web Services","AWS Step Functions","Amazon Bedrock AgentCore Gateway","Model Context Protocol","Amazon Bedrock","Agent Toolkit for AWS","Kiro"]
related_patterns: ["agent-protocols-become-interoperability-layer","agent-ready-software-exposes-capabilities","harness-architecture-outlives-model-choice"]
novelty_label: "new_fact"
what_changed: "AWS now exposes Step Functions workflows as AI agent tools through Bedrock AgentCore Gateway; the prior covers agent skills for Managed Workflows for Apache Airflow."
prior_refs: ["memory:520c4f04-1f0b-8de6-ad5a-b0983856a50f"]
source_url: "https://aws.amazon.com/about-aws/whats-new/2026/10/step-functions-workflows-ai-agent-tools"
source_urls: ["https://aws.amazon.com/about-aws/whats-new/2026/10/step-functions-workflows-ai-agent-tools"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:aws-step-functions-workflows-can-now-be-used-33d8c0"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/aws-step-functions-workflows-can-now-be-used-33d8c0/","markdown":"https://newruntime.com/posts/aws-step-functions-workflows-can-now-be-used-33d8c0.md","json":"https://newruntime.com/posts/aws-step-functions-workflows-can-now-be-used-33d8c0.json"}
---

# AWS Step Functions workflows can now be used as AI agent tools via Amazon Bedrock AgentCore Gateway

## Retrieval answer

AWS has made existing AWS Step Functions workflows available as tools for AI agents through Amazon Bedrock AgentCore Gateway. The change matters because it lets an agent decide whether to initiate a business process while Step Functions executes that process in its predefined order, rather than leaving the model to sequence operational steps itself.

## Summary

AWS has made existing AWS Step Functions workflows available as tools for AI agents through Amazon Bedrock AgentCore Gateway. The change matters because it lets an agent decide whether to initiate a business process while Step Functions executes that process in its predefined order, rather than leaving the model to sequence operational steps itself.

## What happened

AWS announced on October 29, 2026 that customers can use AWS Step Functions workflows as tools for AI agents through Amazon Bedrock AgentCore Gateway. The announcement applies to both existing Standard and Express Step Functions workflows. An agent can decide when to run a workflow and what input to provide; Step Functions then runs every step as the workflow author defined it.

Step Functions is AWS’s visual workflow service for building distributed applications. AWS says it can orchestrate more than 220 AWS services. The new connection places those workflows behind AgentCore Gateway, which connects AI agents to tools through the Model Context Protocol (MCP). In practical terms, a workflow becomes an agent-callable operation rather than a sequence the agent must plan and execute step by step.

AWS’s example is a customer-support agent handling a refund. The agent can start a refund workflow, while the workflow performs the eligibility check, manager approval, and credit step. The announced boundary is important: the agent supplies the decision to invoke the process and the relevant inputs, but the policy sequence is encoded in Step Functions. AWS says this allows multi-step business processes to run in the same order every time.

The capability also covers looking up workflow runs. AWS gives an operations example in which an agent tells an on-call engineer which order-processing runs failed overnight and where they failed. That does not mean the announcement gives the agent authority to repair the failure; it gives it a way to retrieve execution information and point a human operator to the broken step.

The feature is available in all AWS Regions where Amazon Bedrock AgentCore Gateway is available. AWS says customers pay standard rates for Step Functions and AgentCore Gateway; the announcement gives no separate launch price. AWS directs builders to the Step Functions Developer Guide tutorial and to its AgentCore Gateway documentation for supported operations and setup options. It also points users of AI coding assistants such as Kiro to the Agent Toolkit for AWS for skills and guidance.

The public announcement is available at https://aws.amazon.com/about-aws/whats-new/2026/10/step-functions-workflows-ai-agent-tools.

## What the archive adds

This is a new integration event on a known AWS agent stack. Earlier Field Notes examined governed semantic layers for agents on AWS, including [AtScale and Amazon Bedrock AgentCore](https://newruntime.com/posts/build-a-semantic-layer-for-agentic-ai-on-b577b3/) and [Strategy Mosaic and Amazon Bedrock AgentCore](https://newruntime.com/posts/build-a-semantic-layer-for-agentic-ai-on-ec1cc0/). Those notes concerned governed access to enterprise definitions and data; this announcement concerns governed execution of operational workflows.

## Trend signal

The announcement supports [Agent protocols become the interoperability layer](https://newruntime.com/patterns/agent-protocols-become-interoperability-layer/). AgentCore Gateway exposes Step Functions through MCP, turning an existing workflow service into a tool surface available to an agent harness. It also supports the related archive assumption that protocol-exposed tools can remain usable as the number of available tools grows, although this announcement supplies no evidence about tool-selection quality or scale.

## Connections to other events

An earlier Field Note covered [Amazon MWAA adding four AI agent skills to the Agent Toolkit for AWS](https://newruntime.com/posts/amazon-mwaa-adds-four-ai-agent-skills-to-93d318/). Together, the two items show AWS extending both sides of its builder experience: skills and guidance for constructing agent-enabled systems, and a gateway path for connecting agents to established workflow automation. The present announcement does not state that MWAA skills are required to use Step Functions through AgentCore Gateway.

## Why it matters

Teams often want an agent to handle the ambiguous front end of work—interpreting a support request, recognizing an exception, or deciding that an investigation is needed—without allowing it to improvise every operational action. This release offers an AWS-native split between those responsibilities. The model can select a bounded workflow and pass inputs, while the workflow preserves ordering, approval steps, and the integrations already encoded by the application team.

That split can make agent adoption easier in processes where consistency and auditability matter more than conversational flexibility. Existing Standard or Express workflows can be reused rather than translated into prompts or rebuilt in a separate agent framework. For operations teams, access to workflow-run information may also make an agent a useful interface for triage. The value depends on the workflow already being a meaningful representation of the organization’s policy; an unreliable or incomplete workflow does not become reliable merely because it is callable through MCP.

## What happens next

- **More AWS services with established control planes will be exposed as bounded agent tools through Gateway integrations.** (6 months, 65%). Watch for: AWS announcements naming additional service connector targets.
- **Teams will use Step Functions as a reviewable execution boundary for agent-initiated operational actions.** (12 months, 60%). Watch for: reference architectures that combine agent invocation, approvals, and workflow-run observability.

## Limits and uncertainty

AWS’s announcement establishes availability, supported workflow types, regional scope tied to AgentCore Gateway, and standard billing treatment. It does not specify latency, throughput, permission-design guidance, reliability under agent-driven load, or which AgentCore Gateway operations are available beyond the linked documentation. It also does not compare this approach with directly exposing individual APIs to an agent, nor does it establish whether agents select the right workflow reliably when many tools are available. The refund and failed-order examples are illustrative, not measured customer results. Teams still need to define authorization, input validation, error handling, and human review for their own workflows.

## Skill to build now

Choose one existing workflow with a clear business contract, such as a refund request, account-change approval, or failed-job investigation. Write down the exact inputs an agent may supply, the approval and authorization checks that must stay in the workflow, and the run information an operator needs back. Then test whether the workflow is safe and intelligible as one bounded tool call rather than a collection of low-level actions. This exercise exposes whether the workflow is genuinely a reusable capability or merely an implementation detail.

## Reader takeaway

AWS has made Step Functions a callable tool layer for AI agents via AgentCore Gateway and MCP. The significant design move is not that an agent can automate a process; it is that the model can initiate a process while a predeclared workflow, rather than the model, controls how the operational steps run.

## Open question

Will AWS provide evidence that agents can select and parameterize large sets of workflow tools reliably, while preserving the permission and review boundaries that make those workflows valuable?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agent Tools](https://newruntime.com/topics/agent-tools.md): Explore the agent-tools topic hub.
- topic hub: [Mcp](https://newruntime.com/topics/mcp.md): Explore the mcp topic hub.
- topic hub: [Skills](https://newruntime.com/topics/skills.md): Explore the skills topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)