AWS re:Invent 2025 - Transforming Supply Chains with Amazon Bedrock AgentCore (API206)

Transforming Supply Chains with Amazon Bedrock AgentCore

Key Challenges in Supply Chain Management

• Volatility, disruption, and transformation are common themes in modern supply chain management
• Supply chain management involves working with rapidly shifting data across many dimensions and sources
• This high-complexity, high-dimensionality scenario is well-suited for AI, especially when combined with domain expertise

Fujitsu's Uvance Decision Intelligence Approach

• Fujitsu's Uvance decision intelligence offerings aim to enable teams through data and AI
• The approach combines Fujitsu's technical expertise with the domain expertise of customers
• This delivers high-quality decisions at pace to create positive business and social impact

Uvance SCM: First Generation Solution

• Fujitsu's first-generation supply chain management (SCM) solution leveraged AI agents to automate tasks and manage disruptions
• For example, the solution could quantify the impact of a 2024 earthquake on supply chains and begin mitigation within 2 hours, while competitors took 3 days
• However, the solution faced technical challenges, particularly around agent drift, where agents would move away from the business context over time

Addressing Agent Drift with Amazon Bedrock AgentCore

• Fujitsu addressed the agent drift challenge by integrating Amazon Bedrock AgentCore into their Uvance SCM solution
• Bedrock AgentCore provides a single, coherent platform for managing AI agent lifecycles, replacing the previous "mishmash of components"
• Key Bedrock AgentCore components:
  • AgentCore Runtime: Provides the runtime environment for all AI agents
  • AgentCore Gateway: Handles integration of supply chain data sources and tools
  • AgentCore Identity: Manages centralized access control for agents
  • AgentCore Observability: Provides deep insights into agent performance

The Guardian Agent: Preventing Agent Drift

• Fujitsu created a "Guardian Agent" to monitor and correct agent drift
• The Guardian Agent uses AgentCore Observability data and supply chain manager feedback to identify cases where agent responses are degrading
• It then formulates updates to the agent prompts to correct the drift, storing the updates in a central DynamoDB table
• This creates a human-centric approach to evolving the AI agents based on supply chain manager input

Business Impact and Results

• Uvance SCM has enabled customers to see meaningful boosts in SCM team efficiency, leading to significantly faster responses to supply chain incidents
• Integration of new data sources has been accelerated by around 10x thanks to Bedrock AgentCore
• These benefits improve supply chain resilience against both everyday issues and major adverse events

Example Use Case

• A customer faced the challenge of reducing required resources, as their most labor-intensive task was manual inventory ordering
• Using Uvance SCM, they were able to reduce the number of workers required for this task from 50-100 down to just 10
• The solution's machine learning-based predictions helped the customer avoid lost sales and reduce associated costs

Key Takeaways

• Fujitsu's Uvance Decision Intelligence approach combines technical expertise with domain knowledge to deliver high-quality, AI-powered supply chain solutions
• Amazon Bedrock AgentCore provided a critical platform for managing AI agent lifecycles and preventing agent drift, a key technical challenge
• The "Guardian Agent" concept demonstrates a human-centric approach to evolving AI systems based on user feedback
• Uvance SCM has delivered significant efficiency gains and improved supply chain resilience for customers, with one example showing a 5-10x reduction in required resources for a key task