by Julien Debard and Edward Tavares
The electric utility industry stands at a critical inflection point. With unprecedented demand growth of 2.5% annual growth forecasted through 2035 or five times the 0.5% annual growth rate of 2014-2024 (Bank of America data) and coupled with 104 GW of power generation scheduled for retirement by 2030 while only 22 GW of firm replacement capacity currently planned (The Department of Energy’s July 2025 Resource Adequacy Report), Utilities face operational challenges that manual processes simply cannot address at scale. The Department of Energy's stark warning that blackouts could increase by 100 times by 2030 under the current trends underscores the urgency for transformative solutions (U.S. Department of Energy Report: “Evaluating the Reliability and Security of the United States Electric Grid”).
In this context, and as utilities face the convergence of extreme weather, regulatory upheaval such as the One Big Beautiful Bill Act, and exponential demand from data centers and electric vehicles, the question is no longer whether to adopt AI agents but how rapidly to implement them to safeguard grid reliability and public trust.
Utilities are confronting a perfect storm of operational stressors. Global electricity demand jumped 4.3% in 2024, its fastest peacetime growth on record (IEA, 2025). Data centers alone may consume 12% of all U.S. electricity by 2028 (DOE). Electric vehicles are projected to increase global demand sevenfold by 2030 (IEA Global EV Outlook 2024). At the same time, the sector’s aging infrastructure is being stretched thin. One hundred and four gigawatts of generation will retire by 2030, yet only a fraction is slated for reliable replacement. Maintenance demands are rising as transmission and distribution grids, often built decades ago, face mounting stress.
The challenge is compounded by the climate. Weather-related events now cause 80% of major U.S. outages. Hurricane Helene alone triggered 431 transmission failures in 2024, the highest ever for a single event, with $27 billion in weather-related damages that year (NERC, 2025).
Policy adds another layer of urgency. The OBBBA, enacted July 4, 2025, compressed timelines for renewable deployment and removed key incentives. Utilities are being forced to rapidly adjust both strategy and investment. The sector must pivot from manual, legacy processes built for predictable, centralized generation to agile, data-driven operations fit for today’s volatility.
Grid operators face fundamental data and AI-related challenges that limit their ability to manage modern grid complexities. Utility data remains trapped in isolated systems across departments and vendors, creating a fragmented operational picture. Advanced Metering Infrastructure readings stored in vendor-specific NoSQL databases cannot be joined with outage logs in legacy Geographic Information Systems, while pole-inspection reports saved as photographs, Word documents and Excel files on local servers create barriers to comprehensive analysis.
This fragmentation is exacerbated by inconsistent data formats and poor governance. Field crews may report asset inspections using mobile devices tied to a SQL database while SCADA time-series data arrives in CSV files with non-standardized timestamps. Such inconsistency leads to unreliable forecasts and less-than-optimized maintenance schedules that critically affect grid stability and reliability.
Legacy data warehouses simply cannot accommodate the influx from distributed energy resources. The proliferation of rooftop solar, streaming second-by-second data, can overwhelm traditional systems and impede the real-time insight needed for anomaly detection and rapid response. As the grid grows more distributed and complex, these limitations become existential.
AI agents represent a step-change in operational capability. Unlike conventional automation, which executes static rules, AI agents can synthesize massive and diverse datasets, learn from outcomes, and make context-aware decisions across domains such as asset management and outage response. These systems augment, rather than replace, human expertise.
As Julien Debard, Director for Energy and Utilities at Databricks, presents it, “The key to successful AI agent deployment lies in recognizing that full automation isn't the immediate goal: intelligent augmentation is”. This is acknowledging that while grids are operated too manually today to effectively manage energy mix complexity, load forecasting, and outage response, jumping to full automation would be premature and potentially dangerous. Instead, the evolution will follow a clear progression where AI agents provide recommendations that human operators must approve, to exception-based control where agents handle routine decisions autonomously but flag unusual situations for human review, and finally to autonomous operations within defined parameters.
Hawaiian Electric (HE) experience with AI agent implementation provides a practical roadmap for utilities considering this transformation. The journey began with a clear business problem: “Modern utilities operate incredibly complex systems in demanding regulatory environments. Our regulators, customers and other stakeholders frequently ask us questions and they deserve accurate and thorough responses. Accomplishing that is not a simple task. Employees spend significant time researching and reviewing past regulatory filings and other operational data to develop our response. We need to use AI to comb through our files to find relevant sources and citations that our employes can use to develop responses.” – Edward Tavares, Chief Information Officer at HE.
The traditional process was both inefficient and error-prone, requiring manual searches through thousands of documents, time-consuming cross-referencing across multiple regulatory sources, and creating inconsistent responses to information requests while making it difficult to scale expertise across the organization.
Working with Databricks, HE developed a Retrieval Augmented Generation (RAG) model proof of concept that transformed regulatory document querying. The implementation leveraged Databricks Vector Search for semantic search across regulatory documents, Unity Catalog for centralized governance and security controls, and Lakeflow Declarative Pipelines for consistent data preparation and availability.
The results exceeded expectations. Query response times dropped from five minutes to five seconds—a 60X improvement—while the entire system was implemented in just two weeks. The conversational RAG chatbot now acts as a source of truth for legal teams by providing specific page references for every response, building trust through transparency and allowing users to verify AI-generated insights against original sources.
This transparency proves crucial for regulated utilities where accuracy and auditability are paramount. The success of this initial deployment laid the foundation for expanding AI capabilities across other departments and operational areas.
This shows how a focused approach on a defined use case can be solved quickly and efficiently. It is by building such agent one brick at a time that Utilities will get true value from today’s data and AI solutions.
The transformation potential of AI agents extends across multiple operational domains, with four use cases demonstrating particular promise for grid operations.
Predictive Asset Management represents perhaps the most immediate opportunity. Traditional scheduled maintenance based on time intervals often leads to surprise failures and unnecessary work orders. AI agents can continuously analyze asset health data, identifying equipment degradation patterns and optimizing maintenance schedules to prevent failures before they occur. One example of such solution, built on Databricks, enabled the real-time monitoring of 1.5 million customers, 2,400 substations, and 250,000 devices. The system shifted from monthly to near real-time reporting, significantly improving reliability metrics while reducing maintenance costs. Another example of Data and AI used to prevent incidents and increase safety is in wildfire prevention. Utilities have been able to combine advanced geospatial analytics, satellite images, and LiDAR data, and processing terabytes of weather data to enable faster, more accurate risk insights. In a particular case, the initiative increased power outage data analysis coverage by 3.3X, improved accuracy by 4.1X, and reduced processing time by 64X.
Intelligent Outage Response addresses one of Utilities' most visible challenges. Manual crew dispatch based on phone calls and paper processes leads to delayed restoration and suboptimal resource allocation. AI agents can process outage events alongside weather-, workforce-, and inventory-data to optimize crew deployment and reduce restoration times.
Dynamic Grid Management tackles the growing complexity of renewable integration with legacy power generation plants and of distributed energy resources. Static load forecasts and manual renewable intermittency management create inefficiencies and reliability challenges that compound as renewable penetration increases. For example, the integration of renewable energy sources introduces unprecedented frequency management challenges that legacy grid operations cannot adequately address. Traditional fossil fuel generators provided natural inertia through massive rotating turbines that helped stabilize grid frequency, but wind and solar installations lack this mechanical inertia, creating voltage and frequency fluctuations that require constant balancing. This challenge is compounded by new demand patterns from AI and data center workloads that can instantly spin up thousands of GPUs in milliseconds, creating sudden power draws that legacy frequency regulation systems cannot anticipate or accommodate. AI agents can process real-time frequency data, weather forecasts, and computational demand signals simultaneously to predict and preemptively adjust grid operations, maintaining stability across a network that now includes both intermittent generation and volatile AI-driven consumption that change faster than human operators can respond.
Strategic Investment Planning enables data-driven capital allocation decisions. Siloed CAPEX decisions based on static forecasts often underdeliver on expected returns, particularly as demand patterns shift rapidly. AI agents can integrate asset health data, EV adoption trends, and development plans to rank and simulate CAPEX scenarios, helping utilities deliver maximum returns while future-proofing their networks against evolving demand patterns.
Successful AI agent deployment requires a robust technical foundation that addresses the data challenges plaguing traditional Utility systems. A modern data platform must handle multimodal data types—structured operational data, geospatial information, images, audio, video, and unstructured documents—within a single, governed environment. Databricks' lakehouse architecture provides this foundation by combining the flexibility of data lakes with the performance and reliability of data warehouses.
Avoiding vendor lock-in requires platforms built on open-source foundations. Delta Lake, Unity Catalog, and open data formats ensure long-term flexibility while enabling integration with existing systems.
For Utilities handling sensitive customer data and critical infrastructure information, governance capabilities are non-negotiable. Unity Catalog provides fine-grained access controls, audit logging, and lineage tracking across all workspaces, ensuring that AI agents operate within appropriate security boundaries while maintaining compliance with regulatory requirements.
Modern grids require sub-second decision-making capabilities. Structured Streaming, Lakeflow Pipelines, and real-time model serving enable AI agents to process high-velocity data streams and provide immediate insights for operational decision-making, bridging the time gap between data ingestion and actionable intelligence.
Utilities considering AI agent deployment face several common challenges that can be addressed through proper planning and partnership. Many Utilities lack in-house expertise for advanced AI system development, making partnerships with experienced vendors essential while investing in workforce development to build the necessary long-term organizational capability.
Legacy systems integration requires careful planning and phased approaches, ensuring that critical applications that have not yet been modernized can operate on the new set of unified data on the cloud.
Security and compliance concerns demand robust governance frameworks and security controls that address regulatory requirements; all while enabling innovation. Unity Catalog's comprehensive security model demonstrates how modern platforms can meet Utility-grade security requirements without sacrificing functionality.
Cultural change management may prove the most challenging aspect of AI agent deployment. The transition from manual to AI-assisted operations requires significant organizational transformation. Clear communication about AI's role as an augmentation tool rather than replacement, combined with extensive training and gradual implementation, helps build organizational acceptance and ensures successful adoption.
The transition to autonomous grid operations requires careful orchestration to build organizational trust and ensure system reliability. This evolution typically unfolds across three distinct phases, each building confidence and capability for the next level of automation.
The initial phase focuses on human supervision, where AI agents provide recommendations with full transparency while human operators review and approve all actions. During this period, detailed logging of decisions enables continuous improvement while focusing on non-critical applications to build confidence. Success metrics include accuracy of AI recommendations versus human decisions, time savings from faster information processing, and user adoption rates.
The second phase introduces exception-based control, where AI agents handle routine decisions autonomously but complex or unusual situations automatically escalate to humans. This phase emphasizes continuous learning from human feedback while gradually expanding to more critical applications. Key metrics shift to percentage of decisions handled autonomously, reduction in false positive escalations, and improvement in response times.
The final phase enables autonomous operations within defined parameters, with human oversight focusing on strategic guidance and exception handling. Continuous monitoring and adjustment of operational boundaries ensure safe expansion while full integration with enterprise systems maximizes value. Success measures include overall operational efficiency improvements, customer satisfaction scores, and cost reduction achievements.
Performance improvement: The financial case for AI agent deployment becomes compelling when considering the scale of Utility operations and the cumulative impact of incremental improvements. Outage response optimization can reduce service restoration time by 30% through improved crew dispatch, while predictive asset management can deliver 25% improvement in maintenance efficiency. Load forecasting accuracy improvements of 15-20% translate to better resource planning and reduced operating costs, while customer service automation can cut call center costs by 40-50%.
Cost Avoidance: Beyond operational efficiency gains, AI agents enable significant cost avoidance through earlier detection of equipment issues that prevent costly emergency repairs, optimal resource allocation that reduces operational costs, and faster, more accurate regulatory reporting that reduces compliance costs and penalties.
Revenue enhancement opportunities include improved grid reliability that translates directly to customer satisfaction and retention, better load management that enables optimized participation in energy markets, and enhanced service quality that supports customer satisfaction in competitive markets.
The transformation to intelligent grid operations won't happen overnight, but Utilities that begin their journey today will be best positioned to navigate the challenges ahead. Success requires identifying priority low-risk, high-value use cases with experienced technology partners to achieve faster time-to-value, while planning for scaling successful pilots across the organization.
The Utility industry's transformation from manual to autonomous operations represents one of the most significant technological shifts since the development of the electric grid itself. While the challenges are substantial—from aging infrastructure to extreme weather events to unprecedented demand growth—the tools to address them are available today.
AI agents offer a path forward that amplifies human expertise rather than replacing it, building trust through transparency and gradually expanding capabilities as organizations become comfortable with intelligent automation. Hawaiian Electric's success in reducing regulatory document query times from five minutes to five seconds in just two weeks demonstrates that the technology is ready for production deployment.
The question facing Utility leaders isn't whether AI agents will transform grid operations—it's whether their organizations will lead this transformation or struggle to catch up. The next 18 to 24 months will likely determine which Utilities emerge as leaders in the Age of Electricity.
The age of the intelligent grid has begun. The utilities that embrace AI agents today will power the communities of tomorrow.
Visit the Databricks for Energy solution page to learn more.
Subscribe to our blog and get the latest posts delivered to your inbox.