1. Deployed, Not Embedded: Samta Kapoor (EY) opened with a tension: 72% of energy leaders are prioritizing AI governance, up 25 points since 2024, even as 98% rethink their AI budgets under cost pressure. Her ROI heatmap across marketing, finance and R&D showed most initiatives running negative, the gap between bolt-on AI, which speeds up existing processes, and built-in AI, which redesigns outcomes so people and agents work together from the start. Her takeaway: the industry needs business redesign, not more tools.​
​
2. From AI Pilots to Production: Mark Lesiw (Xcel Energy) defines production as five tests, in use, valuable, safe, reliable, owned, not a launch date. His data: roughly 3,000 employee-built agents unlock nearly twice the hours his central team builds directly. His thesis: build the builders, govern the platform, and let proven local wins become the enterprise standard.
 
3. When Prediction Becomes Action: Marcus Johansson (AEP) used AEP’s three-pillar wildfire framework to show where AI acts today, satellite and drone imagery turned into prioritized work orders, cameras and forecasts feeding a digital twin, and where it doesn’t: emergency response stays a human call. His line: miles scanned is activity, fewer ignitions is outcome.

4. What Makes Energy AI Investable: Brandon Cavallaro (Quantum Capital Group) brought the investor’s lens. AI projects fail on data readiness, a technology-first mindset and poor change management, not technology, citing 95% of generative AI pilots failing to show ROI. Quantum’s own discipline, define the opportunity, benchmark, kill weak proofs early, has produced 150+ workflow accelerators across its portfolio.

5. Before AI Comes Asset Intelligence: Ken Chadha (Alectra Utilities) showed how fragmented systems, GIS, SCADA, ERP, each right for their own purpose, leave no complete picture of the grid. Alectra’s Integrated Network Management program doesn’t replace those systems, it reconciles them, feeding corrected data back so AI builds on trusted inputs instead of amplifying their gaps.

6. From Promise to Proven Returns: Adrian Rodriguez-Herrera (SLB) closed with SLB’s AI portfolio across the upstream lifecycle, 200+ solutions from its seven Innovation Factori centres, running on a shared foundation model. His point: industrial-grade AI takes years of engineering, and the payoff is a model still shaping real decisions long after launch.

The Results: Three threads ran through all six talks. First, the human-AI boundary: where AI recommends, acts or defers to a person is a deliberate design choice, made explicit in AEP’s framework. Second, evidence over enthusiasm: Samta’s heatmap, Marcus’s activity-versus-outcome distinction, Brandon’s investor filter and Mark’s rule that every solution needs an owner and a value case all pointed the same way. Third, repeatability: Mark’s operating model, Ken’s data foundation and Adrian’s models built over years all turn one success into many, scaling, as Marcus put it, is about what already works, not adding more tools.

Deployment is not the finish line, embedding is. What separates the companies that get there is rarely the algorithm. It’s the data they trust, the decision rights they’ve clarified, the evidence they can show, and the operating model that keeps learning.