AWS re:Invent 2025 - Build and scale AI: from reliable agents to transformative systems (INV204)

Building Trusted AI Agents at Scale

Importance of Trust in AI Systems

• Building AI agents that can be trusted for production systems is a key challenge
• Lack of reliability, transparency, and safety can turn the most brilliant algorithms into expensive experiments
• Trust must be built into AI systems from the ground up

Key Pillars of Trusted AI Agents

1. Reliability:

   • Importance of building on a secure, extensive, and reliable global cloud infrastructure
   • AWS offerings like EC2 instances, Trainium chips, and co-designed hardware/software for faster, safer, and more efficient AI workloads
   • Customization capabilities to align AI models with business needs and domain-specific requirements

2. Transparency:

   • Importance of observability and visibility into AI agent behavior, performance, and decision-making
   • Amazon SageMaker HyperPod's built-in observability for ML infrastructure and workflows
   • Agent Core observability to trace agent actions, replay workflows, and audit agent behavior

3. Safety and Governance:

   • Importance of setting clear guidelines and policies for AI agent behavior
   • AWS Responsible AI Lens to guide best practices for secure, compliant, and ethical AI deployment
   • Agent Core's identity management, sandboxing, and policy controls to enforce data and access restrictions

4. Ease of Use:

   • Importance of making AI agent development accessible to a wide range of users
   • Strands: an open-source, model-driven framework for building and running AI agents
   • Genai Innovation Center to help organizations move from prototypes to production-ready AI systems

Real-World Examples and Impact

1. Cohere Health:

   • Built an agentic system using Bedrock and Agent Core to automate medical coverage reviews
   • Achieved 30-40% faster reviews with fewer errors, providing faster answers to patients

2. Lyft:

   • Transformed customer support experience with AI-powered intent detection and resolution
   • Achieved 55% of customer interactions resolved without human agents
   • Reduced average resolution time from 16 days to under 3 minutes

Key Takeaways

• Trust is essential for scaling AI agents in production environments
• AWS provides a comprehensive set of tools and services to build reliable, transparent, safe, and easy-to-use AI agents
• Successful AI agent deployment requires aligning technology capabilities with business needs and user experience
• Partnerships and collaboration are crucial for overcoming challenges and driving real-world impact