🦄 Making great presentations more accessible.
This project enhances multilingual accessibility and discoverability while preserving the original content. Detailed transcriptions and keyframes capture the nuances and technical insights that convey the full value of each session.
Note: A comprehensive list of re:Invent 2025 transcribed articles is available in this Spreadsheet!
Overview
📖 AWS re:Invent 2025 - Transforming Supply Chains with Amazon Bedrock AgentCore (API206)
In this video, Chris Bingham from Fujitsu presents their UVANCE SCM solution for supply chain management, addressing challenges of volatility, disruption, and transformation using AI agents built on AWS Bedrock Agent Core. The solution helped a customer quantify earthquake impact within 2 hours versus competitors' 3 days. A key innovation is the Guardian Agent that combats agent drift by combining Agent Core observability data with supply chain managers' feedback to automatically update prompts. The system reduced workforce needs from 1,000 to 500 people, achieved over $10 million in cost savings, and generated accurate prediction models for 1.2 million SKUs with less than 10% error rates. Bedrock Agent Core accelerated new data source integration by roughly an order of magnitude.
; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.
Main Part
Supply Chain Challenges and Fujitsu's First Generation AI Solution
Hello, welcome everybody. Thank you for coming. I'm Chris Bingham, AWS Ambassador and Country CTO for Switzerland at Fujitsu. At Fujitsu, we're applying AI to real business issues that go beyond just code generation, which we've heard a lot about today and this week at re:Invent, with the aim of aiding and accelerating skilled people such as supply chain managers.
What challenges do supply chain managers face? Well, not being one myself, I did the 2025 thing and asked an AI, and it told me all of this. The answer is lots and varied challenges. But I feel they can be summarized with three key words: volatility, disruption, and transformation. Is anybody else experiencing those things in their work these days? It's a bit of a common theme, isn't it?
Overall, supply chain management involves working with a lot of rapidly shifting data across many dimensions and sources, from suppliers, from customers, from customs and regulators, from manufacturing, through to sales and every business unit in between. This kind of high complexity, high dimensionality scenario is exactly where AI excels, particularly where AI works with experienced professionals and domain experts who can wield AI to achieve positive business impacts.
Creating positive impacts through data and AI is at the core of Fujitsu's UVANCE decision intelligence offerings. A key tenet of UVANCE's decision intelligence is to enable your teams through data and AI by providing solutions which act as companions, supporting you in your work through the combination of Fujitsu's technical expertise and your domain expertise. This approach delivers high quality decisions at pace with the aim of creating positive business and social impact.
In this spirit, our teams applied their domain knowledge to create a solution to help supply chain management. Our first generation solution was built to aid supply chain managers through automation and AI. It leveraged AI agents to intelligently help with balancing inventory and, crucially, with managing adverse effects and disruptions. For example, this screenshot shows the supply chain effects of an earthquake which occurred in January 2024. In this real world case, our customer was able to quantify the impact of this earthquake on the supply chains and begin mitigation within two hours. Their competitors, by comparison, had to work around the clock for three days to get to the same point.
However, this version of our solution faced some technical challenges. In a nutshell, speed. Agentic AI, like all other uses of AI, is only effective when it's attuned to reality. Our first generation approach was limited. Lacking a single cohesive platform for managing agent life cycles, the pieced together architecture which we had hindered our ability to adapt agents to changing circumstances. Unchecked, this would lead to agent drift. Agents moving away from the business context over time, resulting in the relevance and quality of their responses degrading, leading to negative impact on the business outcomes our customers could achieve.
Overcoming Agent Drift with Bedrock Agent Core and the Guardian Agent
So how did we build a solution which could rise to this challenge of agent drift? Well, we're at re:Invent, so naturally enough with AWS. Specifically, Bedrock Agent Core provides that single coherent platform for managing our agents' life cycles. Agent Core replaced the mishmash of components at the heart of our AI agents with a single joined up platform. This simplified architecture diagram illustrates how Agent Core integrates into our UVANCE SCM solution. Agent Core Runtime provides the runtime environment for all of our AI agents. Agent Core Gateway handles the integration of all of the supply chain data sources and tools which those agents need, whilst Agent Core Identity manages centralized access control for all of those agents against all of those tools and data sources.
And last but by no means least, Agent Core observability provides deep insights into the how and the why of AI agent performance. That final piece, Agent Core observability, is the linchpin to how we addressed the challenge of agent drift.
By creating yet another AI—yes, more AI on top of AI—the Guardian Agent, as we call it. The Guardian Agent is tasked with guarding against drift. It works to identify cases where our SCM agent's responses have started to degrade and then acts to correct that drift based on both Agent Core observability data and supply chain managers' feedback.
The UVANCE SCM UX includes opportunities for supply chain managers to provide qualitative feedback on the business performance of the agents. This enables UVANCE SCM to capture any shifts in that business performance independently of any technology metrics. The Guardian Agent can then combine that business performance feedback with observability data from Agent Core, most especially the chain-of-thought data like the example which we see here, to work out where in that chain of thought the agent has started to drift.
In doing so, formulate modifications to the prompts used for that agent. These prompt updates are then stored in a central DynamoDB prompt table and utilized the next time that drifting agent is called. This means that automated analysis of and reaction to agent drift is triggered by human feedback. By supply chain managers identifying insufficient business context fit implicitly through their feedback, we create a human-centric approach to evolving the AI agents which are supposed to help the humans.
So with that, I'm now going to show you a quick video demonstration of UVANCE SCM. When the project started, the customer had 1,000 people working under one SCM operation. Their resources were aging, and young veteran workers were expected to retire in three years. Reducing the required resources was one of their most pressing challenges.
Their most labor-intensive task was ordering components. They had about 50 to 100 workers at each plant on this task only because their systems were silent. They had to manually calculate the inventory to identify excess and stock items using Excel sheets. One person could cover up to just 10 items, so they needed 1,000 people to cover their global operation.
So they built this dashboard. Now all they have to do is solve all modifications from the system. This dashboard is now used by the company's top management and also by the procurement team.
Here in this view, you can see the carry train at one, factoring in its procurement, production, and sales. I haven't seen that can do the same. The yellow dot indicates the time for reorder. The red dot indicates stockout, but the purple part shows the opportunity to be lost due to stockout. The blue line shows the optimal ordering according to a dedicated prediction model generated by our evolution.
In just two months, the AI generated accurate prediction models with error rates of less than 10 percent across 4,300 component categories covering 1.2 million SKUs up to 2 million ECAs. This led to more than $10 million in cost savings and also contributed to resource savings from 1,000 to 500. So by just following this blue line, the customer not only reduced the cost associated with the manual costs, but also increased the revenue by avoiding the loss of sales opportunities.
We can aim to have the necessary resources again from 500 FDs to 250 FDAs. In these operation labs, humans take actions by checking automatic alerts from the system. But we are now in the AI age, where further automation for even larger business market is possible. As this view is just for demonstration, state out alerts are rounded up in and using money. But in the actual operation, alerts are automatically processed by those languages. So while those ages are places in the past, this job description defines what the agent is expected to be, and the returns to find what data connects us and what actions it. The selected agents here are moved from left to right in the compound scale as the agents cross them. The orchestrated agent selects the best solution from those agents' answers. These agents are evaluated according to these indicators, and the farmers are trained according to humans. If AI agents, in other words, agents designed with certain songs were implemented after coming sound data works are integrated and streaming, and how to be able to take appropriate actions based on health understanding. This illustrates the concept of Fujitsu devices, which is to add business value from tens of millions to hundreds of millions of dollars through reorganization and export implementation by solving entire problems using the same thing. You can see all your wise hands in the winter. So, for example, in this dashboard, when an earthquake occurs, the system identifies its impact on the customer supply chain. Most companies can manually calculate livestreams using excel sheets, such as which suppliers and components are likely to. But identifying the impact such as which is extraordinary, and we will 202, the peninsula in Japan. This custom was able to ballpark within 2 hours. The officers at Japan's Electronics Manufacturing Industry Association spent 3 days without sleep to do the same thing. Japan is an ethnic farm country, but the same approach applies to many natural disasters, including hurricanes and floods, deal with costs such as the recent declaration of Martian.
To summarize, our customers are seeing meaningful boosts to their SCM team's efficiency from deploying Evan SCM leading to significantly more rapid responses to incidents within their supply chain. And thanks to Bedrock Agent Core's capabilities, we've been able to accelerate the integration of new data sources, such as may come when you have a new supplier, by roughly an order of magnitude. These benefits together improve your resilience, both against everyday issues and major adverse events. And with that, if you'd like to learn more about Evan SCM or any of the other solutions we offer, please do come and visit us, we're at booth 1851, which is roughly in that direction. Or please connect with us online via the QR code here, and we will happily arrange for our experts to come and meet with you after Reinvent at your convenience. Thank you very much for your time and do enjoy the rest of your Reinvent.
; This article is entirely auto-generated using Amazon Bedrock.
Top comments (0)