AWS re:Invent 2025 - Automating Amazon Fulfillment Center Operations with Generative AI (IND393)

Automating Amazon Fulfillment Center Operations with Generative AI

Addressing Manufacturing and Supply Chain Challenges with AI

• The presentation discusses how AI is transforming manufacturing and supply chain operations, providing practical examples of driving business value and increased productivity.
• Key challenges facing the industry include building supply chain resilience, addressing an aging workforce, and mitigating geopolitical risks - all of which require AI-driven re-industrialization.
• The goal is to reduce variation, empower employees, and execute at the speed of business by leveraging new AI tools like automation, assistance, and agents.

Reimagining the Software-Defined Factory

• Traditional manufacturing systems are siloed, making it difficult to leverage data and drive AI-powered workflows.
• The desired future state involves tight IT/OT integration, any-to-any communication, and a shift from systems of record to systems of action.
• AWS's approach focuses on three key layers:
  1. Digitally composed outcomes that redefine processes
  2. An industrial data fabric to contextualize and scale data
  3. Modernizing manufacturing applications and OT security

AI Systems Engineering

• Traditional systems engineering must be reimagined using new AI tools like automation, assistance, and agents.
• The framework considers the risk and complexity of each task, using the most cost-efficient tool - automation for low-risk/low-complexity, human-in-the-loop for high-risk/low-complexity, and agents for low-risk/high-complexity.
• This enables a step-function leap in capabilities and a new baseline for continuous improvement.

Automating Amazon Fulfillment Center Operations

• Amazon has over 3,000 global facilities, aiming to increase throughput by 10% year-over-year through new construction and retrofits.
• The operational readiness testing (ORT) process to verify 10,000+ modules and 200,000+ components is highly manual and time-consuming.

Iora: Intelligent Orchestration and Recognition Architecture

• Iora is a holistic solution to automate the visual inspection and verification of fulfillment center components during the ORT process.
• Key features:
  • Computer vision to detect and identify components (21 items detected vs. 5-10 by humans)
  • Prompt engineering with technical descriptions, rules, and hierarchical relationships to improve accuracy
  • Manual override capability for the human-in-the-loop
  • Optimization techniques like throttling, latency reduction, and prompt caching
• Pilot results:
  • 92% precision and under 5 seconds latency
  • 60% reduction in manual testing time and number of testers
  • $6M annual savings by automating the ORT process

Expanding the Solution

• Moving from 2D static images to 360-degree continuous monitoring using platforms like OpenSpace
• Exploring agent-based AI to automate testing of equipment like conveyors and machinery
• Demonstrating the value of rapid prototyping and cross-functional collaboration to drive innovation at scale

Key Takeaways

• AI-driven re-industrialization is critical to address supply chain challenges and build competitive advantage.
• Reimagining manufacturing systems with tight IT/OT integration and a shift to systems of action enables AI-powered workflows.
• AI systems engineering leverages automation, assistance, and agents to reduce variation, empower employees, and execute at speed.
• Iora demonstrates how generative AI can automate complex visual inspection tasks, reducing manual effort by 60% and delivering significant cost savings.
• Continuous innovation through rapid prototyping and cross-functional collaboration is key to driving AI-powered transformation in manufacturing and supply chain operations.