Ask one question in plain English, get a supply chain answer

Amazon Web Services has published a walkthrough of a supply chain assistant that answers questions across several business systems in one chat. In the demo, a manager checks for disruptions, sees that a halted supplier puts $750,000 and 23 orders at risk, and sends an emergency request for quotes. It takes three prompts and minutes, instead of the usual 45 to 60 minutes.

The problem it targets

Supply chains produce constant signals: orders waiting for approval, shifting lead times, slow ports, drifting prices. The post argues the trouble is not a lack of data. It is the distance between the data and the person deciding. A manager's morning check can span an ERP (the system that tracks inventory and orders), a monitoring tool, procurement, email and a tracker. That means five systems and a dozen or more switches between them.

What the demo shows

The walkthrough follows one disruption from detection to approval in a single Amazon Quick conversation. Quick is the chat tool the business user sees. Behind it, Amazon Bedrock AgentCore connects to the company systems.

• The user asks what supply chain tools are available and sees four business servers listed, already set up.
• A request to check active disruptions returns three, with severity, region and number of affected products.
• Opening the TechComp Production Halt shows $750K at risk, 23 orders and three days of supply.
• Asking for a fix checks approved alternative suppliers and gets a quote that fits the contract.
• The user can also ask for a morning alert, and the same flow can start from a supplier email.

The three prompts were: any critical disruptions affecting top-revenue products, assess the halt and issue an emergency RFQ to approved alternatives, then "Looks good. Send it."

How it stays controlled

The post says answers come from live company systems and grounded documents, not from the AI model's memory. Every change waits for a named human. Automated runs, such as a morning check or a supplier email, post review-ready cards to an Activity Feed for a person to approve.

The connections use the Model Context Protocol, or MCP, an open standard for how AI models use outside tools. Tools are listed in a registry where a publisher submits a record and a curator approves it before anyone can find it. AgentCore supplies the pieces that handle access, rules and running the agent.

What is not covered

The numbers come from a demonstration, not from a customer. The material gives no cost figures and no results from real companies. The post says it shows how to deploy the pattern, which suggests this is aimed at technical teams. It also warns that tool definitions should be treated as contracts, with new versions kept alongside old ones when changes would break things.

Why it matters

For a business owner, the useful idea is the design, not the brand names. One assistant reaches the systems your team already uses, shows its sources, and asks a person to sign off before anything is sent.