AWS re:Invent 2025 - Powering Enterprise AI: Architecting and Governing Agents at Iberdrola on AWS
Source: Dev.to
Overview
AWS re:Invent 2025 – Powering Enterprise AI: Architecting and Governing Agents at Iberdrola on AWS
In this session Sergio Merchan, Unai Bermejo, and Joseph Santamaria discuss how Iberdrola – Europe’s largest energy company serving 100 million customers – is deploying agentic AI across its operations. The presentation covers Iberdrola’s multi‑region Amazon Bedrock deployment (Spain, UK, US, Brazil) that balances data‑sovereignty with agent reusability, three concrete IT‑operations use cases, the underlying architecture (MCP servers, PostgreSQL with vector capabilities, multi‑account strategy), and the benefits of a framework‑agnostic, serverless, auto‑scaling approach. Future challenges around agent reusability and governance are also examined.
Introduction to Agentic AI and Its Transformative Impact on Enterprise Operations
Speakers
- Sergio Merchan – Global CIO, Iberdrola
- Unai Bermejo – Global AI Expert Engineer, Iberdrola
- Joseph Santamaria – AWS Technical Lead for Energy Customers
The session begins with an overview of AWS solutions for agentic AI, followed by a discussion of why generative AI matters for the energy sector, Iberdrola’s strategy, and detailed use‑case walkthroughs.
What Is Agentic AI?
Agentic AI refers to autonomous systems built around agents that can plan, reason, and adapt to changing conditions. Unlike traditional AI models that produce static outputs, agents can execute multi‑step workflows, incorporate business logic, and interact with humans or other systems in a way that mimics human decision‑making.
Key characteristics
- Autonomous planning – agents decide the sequence of actions needed to achieve a goal.
- Dynamic reasoning – they adjust behavior based on new data or context.
- Workflow orchestration – agents can coordinate with other agents or services to solve complex problems.
This capability enables automation of processes that were previously too intricate for conventional AI, impacting everything from customer interactions to internal operations.
Adoption Trends
- Gartner predicts 33 % of all applications will incorporate agentic AI by 2028.
- 15 % of corporate decisions are expected to be automated through agents within the same timeframe.
The rapid adoption is driven by dramatic productivity gains (up to 10×) and cost reductions (projects that once cost millions now cost thousands).
AWS AI Stack: From Applications to Infrastructure for Building Intelligent Agents
AWS provides a three‑tier AI stack that supports the entire lifecycle of agentic solutions.
Applications (Top Tier)
- Kiro – code‑assistant for developers, accelerating the software development lifecycle.
- AWS Transform – modernizes applications (e.g., .NET → Java) and keeps tech stacks up‑to‑date.
- Amazon Connect – cloud‑based contact‑center solution for multi‑channel customer interactions.
Platform Services (Middle Tier)
- Amazon Bedrock – foundation models (Claude, etc.) and the AgentCore runtime for building agents.
- LangGraph – orchestration framework that enables complex reasoning flows.
- MCP (Managed Compute Platform) servers – serverless compute with auto‑scaling and zero cold starts.
Infrastructure (Bottom Tier)
- Multi‑account strategy – isolates business units while enabling cross‑region scalability.
- PostgreSQL with vector extensions – stores embeddings for semantic search and retrieval.
- Multi‑region deployment – ensures data sovereignty (Spain, UK, US, Brazil) and low‑latency access.
Use Cases Highlighted by Iberdrola
- ServiceNow Change‑Request Assistance – an agent that guides users through change‑request creation, reducing manual effort.
- Networking Incident Enrichment – agents automatically gather context, correlate logs, and suggest remediation steps.
- Template‑Selection Chatbot – conversational interface that helps engineers pick the right configuration templates.
All three are built on Amazon Bedrock AgentCore, leverage Claude models for natural‑language understanding, and store contextual data in the vector‑enabled PostgreSQL layer.
Benefits Realized
- Framework‑agnostic development – teams can choose the best tools without vendor lock‑in.
- Cost‑effective serverless compute – pay‑as‑you‑go with automatic scaling eliminates over‑provisioning.
- Zero cold starts – MCP servers keep agents warm, delivering sub‑second response times.
- Business isolation – multi‑account architecture protects data and simplifies governance.
Future Challenges
- Agent reusability – creating modular agents that can be shared across domains while respecting data‑privacy constraints.
- Governance and compliance – establishing policies for monitoring, auditing, and controlling autonomous actions.
Iberdrola’s experience illustrates that addressing these challenges early is essential for scaling agentic AI safely across the enterprise.
References
- AWS re:Invent 2025 session video
- Gartner research on agentic AI adoption (2028 outlook)