AWS re:Invent 2025 - Powering Enterprise AI: Architecting and Governing Agents at Iberdrola on AWS

Powering Enterprise AI: Architecting and Governing Agents at Iberdrola on AWS

Overview of Agentic AI

• Agentic AI represents the ability to create autonomous systems based on intelligent agents
• Agents can automate complex workflows and tasks in a human-like manner, with the ability to plan, reason, and adapt to changing conditions
• Agentic AI is expected to have a transformative impact on every business process, customer interaction, product, and service

AWS AI Stack for Agentic AI

• AWS offers a comprehensive AI stack to enable agentic AI solutions:
  • Applications: Managed services like Kira (code assistant), Transform (modernization), and Amazon Connect (contact center)
  • Agentic AI Platform: AWS Bedrock, including Agent Core for managing agent-based workflows
  • Infrastructure: AWS Forge for fine-tuning and training custom language models on enterprise data

Iberdrola's Agentic AI Strategy

• Iberdrola is a global energy leader, serving over 100 million customers across Europe, the US, Brazil, and other regions
• Iberdrola's goal is to put AI at the center of their business transformation, reimagining and automating processes across the value chain
• Key success factors:
  1. Standardized Competency Centers: Centralized teams with expertise in architecture, engineering, operations, security, etc.
  2. Agentic AI Platform: Leveraging AWS Bedrock and other services to build a secure, observable, and governed platform
  3. Regional Deployment: Deploying agents in a multi-account, multi-environment strategy to ensure data sovereignty and scalability

Agentic AI Use Cases at Iberdrola

1. Change Request Assistance: Agents assist users in correctly filling out change request templates, reducing delays and errors.
2. Network Incident Enrichment: Agents automate the tagging and resolution of network incidents, improving efficiency.
3. Change Template Recommendation: A chatbot-like agent helps users select the appropriate change template from over 200 options.

Technical Architecture

• Each use case has a dedicated Agent Core runtime, with Docker images deployed through CI/CD pipelines
• Agents consume models through AWS Bedrock's LLM layer, with guardrails to ensure appropriate output
• MCP servers provide database access, and a gateway handles authentication and routing to the appropriate runtime
• Serverless architecture and cost-effective compute with AWS Lambda allow for efficient scaling and utilization

Lessons Learned and Future Considerations

• Challenges in reusing and sharing agents across the organization, leading to duplicated effort and potential inconsistencies
• Strategies to address this:
  1. Ownership and authorization model for sharing agents
  2. Versioning and governance of agents in a central registry
  3. Ability to fork and adapt agents for specific use cases, while leveraging the core functionality

Key Takeaways

• Iberdrola is pioneering the use of agentic AI to transform their business, automating complex workflows and tasks across the energy value chain
• AWS Bedrock and the broader AWS AI stack provide a comprehensive platform to build, deploy, and govern agentic AI solutions
• Careful architectural design, regional deployment, and a centralized competency model are critical to scaling agentic AI initiatives
• Reusability and sharing of agents are key challenges that require a combination of technical and organizational strategies