AWS announced on October 29, 2026 that Step Functions workflows can now be used as tools by AI agents through Amazon Bedrock AgentCore Gateway. The agent decides when to run a workflow and what to send it; Step Functions then runs every step exactly as you defined it. It is a neat answer to the biggest worry about agents in business: what if they skip a step?
Coffee summary
- FACT: Existing Standard and Express Step Functions workflows can be exposed as agent tools.
- FACT: AgentCore Gateway connects agents to tools using the Model Context Protocol (MCP), an open standard for plugging tools into AI models.
- FACT: Agents can also look up past workflow runs, for example to report which ones failed overnight.
- FACT: Available in all Regions where AgentCore Gateway is offered, at standard Step Functions and Gateway rates.
- OPINION: This is the right split of work: AI handles judgment, fixed workflows handle rules.
Quick definitions
- Step Functions: an AWS service that chains steps (over 220 AWS services) into a visual workflow that runs the same way every time.
- AgentCore Gateway: AWS's hub that lets AI agents discover and call tools.
- MCP: the Model Context Protocol, a common "plug" format so many AI apps can use the same tools.
How it works, with AWS's example
A customer support agent receives a refund request. Instead of improvising, it calls your refund workflow. Step Functions then runs the eligibility check, the manager approval, and the credit, in order. Every refund follows policy, no matter how the conversation went.
Second example: an operations agent can check workflow history and tell the on-call engineer which order runs failed overnight and at which step, so they can go straight to the fix.
Why it matters
| Agent alone | Agent + Step Functions |
|---|---|
| Might skip or reorder steps | Steps always run in the defined order |
| Hard to audit what happened | Each run is recorded in Step Functions |
| Policy lives in a prompt | Policy lives in a workflow you control |
Who should care
- Teams already on AWS with approval, refund, onboarding, or data-processing workflows.
- Developers building agents who need guardrails that legal or finance will accept.
- Ops teams who want an agent to summarize failures without giving it free rein.
Limitations
- You still need AgentCore Gateway set up, and you pay for both services.
- The agent still chooses *when* and *with what input* to run a workflow, so input validation inside the workflow remains essential.
- This is AWS-only; it does not help if your processes live elsewhere.
What to do next
Pick one existing workflow with a clear business rule, such as refunds under a set amount. Follow the tutorial in the Step Functions Developer Guide, connect it through AgentCore Gateway, and test with a small internal agent before customers ever touch it.
AIImpish Take
The safest agents are the ones that call trusted processes instead of inventing them. AWS just made that pattern easy for anyone already using Step Functions. If your company is nervous about agents, this is a sensible place to start.
*Sources: AWS What's New, Oct 29, 2026.*
AIImpish