Iberdrola, one of many world’s largest utility corporations, has embraced cutting-edge AI know-how to revolutionize its IT operations in ServiceNow. Through the use of totally different agentic architectures, Iberdrola has remodeled the way in which 1000’s of change requests and incident tickets are managed, streamlining processes and enhancing productiveness throughout departments.
Via its partnership with AWS, Iberdrola applied these brokers in a groundbreaking resolution utilizing Amazon Bedrock AgentCore, concentrating on three key areas: optimizing change request validation within the draft part, enriching incident administration with contextual intelligence, and simplifying change mannequin choice utilizing conversational AI. These improvements cut back bottlenecks, assist groups speed up ticket decision, and ship constant and high-quality information dealing with all through the group.
Amazon Bedrock AgentCore helps Iberdrola deploy production-ready AI brokers seamlessly. With serverless compute capabilities, sturdy safety, and built-in observability, the platform helps Iberdrola scale options throughout departments whereas adhering to enterprise-grade reliability and compliance requirements.
Challenges with change and incident administration
Iberdrola has simplified the multi-phase strategy of change administration utilizing AI-powered validation. A bunch of orchestrated brokers make sure that requests align with supposed modifications whereas formatting and verifying necessary fields in actual time. This optimized method avoids guide resubmissions and drastically reduces processing instances, serving to groups give attention to driving impactful outcomes.
Utilizing a swarm of brokers to carry out contextual enrichment, Iberdrola’s networking division now processes incidents sooner and with better precision. This enrichment lets technicians entry configuration merchandise particulars, overview associated historic incidents, and categorize tickets by surroundings and alert varieties, enhancing response instances and enabling groups to swiftly tackle essential points.
Answer overview
Iberdrola establishes its agentic AI apply by means of a layered structure that separates operational issues whereas enabling seamless integration throughout IT workflows. ServiceNow serves as the first enter supply, and a MicroGateway gives clever routing to direct requests to related brokers. A devoted information layer maintains enterprise data, processing uncooked ServiceNow information by means of extract, remodel, and cargo (ETL) pipelines for agent consumption.
The structure includes three layers:
- Agentic AI assets – This layer encompasses all agent deployments, Mannequin Context Protocol (MCP) servers for standardized information entry, authentication mechanisms, and reminiscence objects that keep contextual data. The design permits domain-specific agent improvement whereas sharing widespread infrastructure providers.
- Inference layer – A streamlined abstraction gives giant language mannequin (LLM) inference capabilities from the group’s portfolio of built-in fashions. This layer gives constant mannequin entry patterns whereas supporting experimentation with out requiring agent modifications.
- Information layer – A complete data basis comprises operational information, analytical datasets, and transactional information. This layer enriches agent capabilities by offering entry to historic patterns, real-time operational standing, and contextual data vital for clever decision-making.
This design permits three distinct use circumstances that tackle totally different operational challenges:
- Enhanced change administration validation – The primary implementation helps the draft part of Iberdrola’s change administration course of by means of a deterministic agentic workflow. A number of specialised brokers work in sequence to validate change mannequin appropriateness and confirm that necessary fields comprise accurately formatted data. When validation errors are detected, the system gives clear suggestions to requesters earlier than permitting development to subsequent phases.
- Clever incident enrichment – The incident administration resolution demonstrates multi-agent orchestration for Iberdrola’s Networking division. A grasp agent receives every incident and selectively engages specialised brokers for tagging, contextual enrichment, similarity detection, and alter affect evaluation. This adaptive method assists technicians by categorizing incidents, figuring out associated historic circumstances, and extracting configuration merchandise particulars.
- Conversational change mannequin assistant – The third use case addresses the complexity of choosing acceptable change fashions by means of a conversational AI assistant. The agent collects details about know-how households, change aims, and deployment environments to suggest appropriate change fashions. The system gives clickable suggestions that open pre-filled change kinds, streamlining the change request course of.
The conceptual structure interprets right into a production-ready implementation by means of Amazon Bedrock AgentCore, which gives managed primitives for constructing and deploying enterprise AI brokers. The serverless method of Amazon Bedrock AgentCore permits Iberdrola to give attention to agent logic relatively than infrastructure administration whereas offering scalability and operational reliability.
Amazon Bedrock AgentCore parts
AgentCore Runtime serves as the muse for agent deployment, accepting containerized brokers constructed with any framework—in Iberdrola’s case, LangGraph—and deploying them by means of Amazon Elastic Container Registry (Amazon ECR) repositories. AgentCore Runtime maintains serverless traits, scaling based mostly on request quantity whereas offering session isolation. Every agent session can run as much as 8 hours for advanced workflows. Logs and metrics generated by AgentCore Runtime are routinely captured by AgentCore Observability. As well as, Iberdrola has configured specific logging to their self-hosted Langfuse occasion for centralized monitoring.
AgentCore Reminiscence gives contextual continuity throughout agent interactions by sustaining reminiscence objects per agent session. Utilizing the reminiscence object, brokers can retailer and retrieve session state, dialog historical past, and intermediate processing outcomes. This functionality is crucial for Iberdrola’s multi-step workflows the place brokers should keep context throughout validation phases or incident enrichment processes.
AgentCore Gateway simplifies software integration by performing as an MCP server that “MCPifies” exterior instruments and providers. Reasonably than requiring customized integration code for every information supply, AgentCore Gateway gives standardized interfaces that brokers can eat constantly. This method is especially invaluable for Iberdrola’s ServiceNow endpoint connections.
AgentCore Id manages each inbound and outbound authentication flows, integrating with Entra ID by means of OAuth 2.0 protocols. For inbound requests, AgentCore Id validates bearer tokens and authorizes entry to underlying assets. For outbound operations, it handles token acquisition and manages safe communication with downstream instruments.
AgentCore Observability captures telemetry information from brokers utilizing OpenTelemetry requirements and surfaces this data by means of Amazon CloudWatch. This integration gives complete monitoring of operational metrics with out requiring extra instrumentation.
Technical implementation
The implementation makes use of LiteLLM as a proxy layer for constant entry to Amazon Nova and Anthropic Claude fashions by means of Amazon Bedrock and varied different fashions. This abstraction permits brokers to work together with totally different mannequin variants utilizing standardized API calls whereas Amazon Bedrock Guardrails gives security controls for mannequin outputs.
The structure addresses Iberdrola’s enterprise safety necessities by means of a digital non-public cloud (VPC) configuration inside AgentCore Runtime, so brokers can securely entry inside assets whereas sustaining community isolation. VPC endpoints present safe communication with inside information sources with out exposing visitors to the general public web.
Customers provoke requests by means of ServiceNow, which communicates by means of a REST API to the MicroGateway that routes requests to acceptable use case brokers. The information structure implements a hybrid method combining real-time operational entry with enriched analytical datasets. Uncooked ServiceNow information flows by means of ETL processes into Amazon Easy Storage Service (Amazon S3) storage, then into Amazon Relational Database Service (Amazon RDS) databases enhanced with pgvector extensions for semantic search.
The logs and metrics generated by the brokers deployed in AgentCore Runtime will be monitored utilizing AgentCore Observability. As well as, Iberdrola makes use of self-hosted Langfuse on Amazon Elastic Kubernetes Service (Amazon EKS) for a holistic view of spans and traces generated by the LLMs and the brokers.
Use case particulars
On this part, we focus on the implementation of two use circumstances talked about earlier: enhanced change administration and clever incident administration.
Enhanced change administration
The primary use case demonstrates an agentic workflow that helps the draft part of Iberdrola’s change administration course of by means of sequential agent execution inside a single AgentCore Runtime. The workflow processes change requests by means of 4 specialised brokers—Rule Extractor, Content material Validator, AIM Mannequin Analyst, and Section Transition—with every agent receiving context from the earlier step.
The implementation consists of the next key parts:
- Single runtime context stream – Brokers function inside one AgentCore Runtime occasion, sustaining seamless context and session state throughout all the validation pipeline
- LangGraph orchestration – Brokers are outlined as a graph construction, enabling visible workflow illustration, conditional branching based mostly on validation outcomes, and complete audit trails
- Vector-enhanced validation – Pgvector-enabled PostgreSQL helps semantic similarity searches, enabling the AIM Mannequin Analyst agent to match change fashions based mostly on technical descriptions relatively than key phrase matching
- Constant processing – Change requests observe equivalent validation steps, assembly compliance necessities and high quality requirements
Clever incident administration
The second use case demonstrates clever multi-agent orchestration for incident administration, the place a Sensible Solver Agent analyzes incoming incidents and selectively engages specialised brokers based mostly on contextual wants. This implementation adapts processing steps to every incident’s distinctive traits, optimizing useful resource utilization whereas offering complete enrichment when wanted.
The implementation consists of the next key parts:
- Clever orchestration – The Sensible Solver Agent analyzes incident content material and determines which specialised brokers to invoke based mostly on lacking context and potential value-add
- Specialised agent engagement – 5 specialised brokers (Tag Classifier, Incident Similarity, Incident Associator, Change Associator, Context Retriever) can be found to offer enrichment based mostly on the element and complexity of the incident
- Adaptive processing – The system adjusts enrichment actions based mostly on incident complexity—easy incidents may solely require tagging, whereas advanced points obtain full contextual evaluation
Classes realized
The implementation of AI brokers at Iberdrola demonstrates how the managed primitives of Amazon Bedrock AgentCore considerably speed up enterprise AI deployment. Amazon Bedrock AgentCore minimized the infrastructure complexity usually required for agentic AI, serving to groups give attention to agent logic whereas attaining scalable and secured cloud assets.“At Iberdrola, we’re extending our manufacturing AI platform with a brand new agentic functionality powered by Amazon Bedrock AgentCore,” says Iñigo Gutierrez, AI International Knowledgeable Engineer at Iberdrola. “Through the use of a managed serverless runtime with built-in identification, reminiscence, and observability, we will ship LangGraph-based brokers that plan, name instruments by means of MCP-style gateways, and function securely inside our VPC. This characteristic strikes us from level automations to reusable, production-grade brokers—lowering engineering cognitive load and accelerating protected supply throughout IT operations.”
Key success elements
The answer provides the next key advantages:
- Function-built runtime – AgentCore Runtime gives a fully-managed fast begin environments to host AI brokers with full session isolation. Moreover, out-of-the-box streaming and MCP and A2A assist from AgentCore Runtime alleviate the necessity to develop customized options and construct assist for these protocols.
- Managed infrastructure – The serverless compute runtimes, identification, and reminiscence providers of Amazon Bedrock AgentCore decrease customized improvement overhead for enterprise-grade capabilities.
- Enterprise safety – VPC assist and complete tagging aligns with stringent IT necessities, accelerating improvement with out compromising safety requirements.
- Open and framework-agnostic – Amazon Bedrock AgentCore suits effectively with improvement tips as a result of you may select the event framework, equivalent to LangGraph, by including a easy decorator. Moreover, it has no restrictions on utilizing third-party or open-source options like Langfuse.
- Scalable software discovery – AgentCore Gateway routinely indexes instruments and gives serverless semantic search, scaling from tens to tons of of targets, completely managed.
Future roadmap
Iberdrola is contemplating the next future enhancements to the answer:
- Agent catalog – Enhance governance and discovery of brokers seamlessly built-in into the Amazon Bedrock AgentCore ecosystem
- New supported protocols and requirements – Evolve Iberdrola’s agent improvement to make use of new protocols supported (equivalent to A2A) by AgentCore Runtime and different managed providers
- Managed orchestration and real-time stream monitoring – Construct platform-provided dashboards that routinely handle and monitor advanced interactions between a number of AI brokers, instruments, or workflows
Conclusion
Iberdrola’s progressive implementation showcases its management and imaginative and prescient in utilizing superior AI applied sciences to remodel its operational workflows. By adopting Amazon Bedrock AgentCore, Iberdrola has demonstrated how organizations can deploy production-ready AI brokers with exceptional effectivity whereas assembly sturdy enterprise safety and scalability requirements. Via its strategic use of Amazon Bedrock AgentCore managed primitives, Iberdrola has realized substantial productiveness good points and unparalleled enhancements in information high quality throughout its change and incident administration processes. This profitable transformation underscores Iberdrola’s dedication to excellence in utilizing clever options to resolve advanced operational challenges. It additionally highlights the distinctive worth proposition of Amazon Bedrock AgentCore: industry-first serverless compute for AI brokers, built-in enterprise-grade safety, and adaptable deployment patterns that accommodate various processing necessities. The platform’s capacity to streamline infrastructure complexity whereas supporting specialised workflows makes it a really perfect basis for enterprise AI initiatives.
Organizations seeking to implement AI brokers in manufacturing environments can draw inspiration from Iberdrola’s architectural patterns and its efficient execution of AI-driven options. Iberdrola’s success serves as a blueprint for accelerating deployments and attaining operational excellence with an Amazon Bedrock AgentCore managed method, which reduces time-to-value and helps the dimensions and reliability demanded by enterprise AI programs.
Concerning the authors
Talha Chattha is a Sr. Agentic AI Specialist SA at AWS, based mostly in Stockholm. With 10+ years of expertise working with AI, Talha now helps set up practices to ease the trail to manufacturing for Agentic AI workloads. Talha is an professional in AgentCore and helps prospects throughout whole EMEA. He holds ardour about meta-agents, async patterns, superior hierarchical options and optimized context engineering for brokers. When not shaping the way forward for AI, he explores the scenic European landscapes and scrumptious cuisines. Join with Talha at LinkedIn.
Unai Bermejo is a International Knowledgeable AI Engineer at Iberdrola. With 10 years of expertise in utilized AI, AI analysis, and software program engineering, Unai now helps Iberdrola set up finest practices and frameworks in AI and agentic initiatives, aligned with company platforms and enterprise wants. He acts as a technical bridge between AI know-how, Cloud engineering groups, and enterprise builders, driving the adoption of scalable, accountable, and excessive‑affect AI options throughout the group.
Xabier Muruaga is the International Head of AI and Information at Iberdrola. With over 15 years of expertise in AI/ML and information‑pushed architectures, he leads the corporate’s technique and governance for safe, cloud‑native, and manufacturing‑prepared AI platforms. His background throughout structure, digital transformation, and vitality applied sciences permits him to drive accountable, excessive‑affect AI and agentic initiatives throughout the group.
Iñigo Gutierrez is a International Cloud AI Engineer at Iberdrola with 5 years of expertise in Cloud structure, platform engineering, and AI enablement. Based mostly in Bilbao, he’s liable for the design, evolution, and governance of the corporate’s company Cloud platforms, making certain they supply a safe and scalable basis for AI and digital transformation initiatives. Iñigo acts as a technical enabler between Cloud engineering groups, AI tasks, and enterprise models, selling standardized practices, operational excellence, and the adoption of accountable, excessive‑affect AI options throughout the group.
