Agent-based AI: a tool for optimised energy management
Agent-based AI is opening up new possibilities for the energy sector and, more broadly, for industry as a whole. Maintenance, production, crisis management, alert prioritisation, automated dispatch: the range of use cases is growing, provided these technologies are grounded in the operational realities on the ground.
In a feature published by Les Échos on 27 April 2026, Marc Allaire, consulting director at BeTomorrow, points out that the challenge is not to imagine an AI capable of managing complex infrastructure on its own. The real value lies instead in identifying controlled use cases where automation can improve service quality, responsiveness and decision-making, without removing the human element from the process.
In the energy sector, as in industry, performance does not rely solely on technical optimisation. It also depends on the ability of solutions to integrate into everyday use, to make information clearer and to facilitate action. For households, for example, the issue is not just knowing when to consume, but making this information understandable, useful and engaging.
“ Optimising energy consumption in households is a problem that has already been well resolved. The real technological leap lies in the user experience and the integration of these solutions. ”
It is with this in mind that we are supporting EDF Solar Solutions in the redesign of its monitoring app. The aim: to transform a data-viewing tool into a more accessible, dynamic and engaging experience for customer-generators. Real-time production visualisation, redesigned usability, and highlighting the tangible impact: the app becomes a tool for user adoption, where technology supports usage rather than complicating it.
Agent-based AI takes on a new dimension when it comes to managing unforeseen situations on critical infrastructure. In the event of an incident, it can help coordinate several successive actions: classifying the event, preparing a response, verifying available information, prioritising alerts, or facilitating the monitoring of operations. But its role is not to replace the human decision-maker. It prepares the ground so that the decision-maker can make decisions more quickly, with information that is better structured, more reliable and more directly actionable.
To achieve this, we favour a multi-agent approach. Each agent is designed to perform a specific task within a controlled scope. This architecture allows complex processes to be broken down into verifiable micro-actions, which limits deviations, reduces the risk of errors and makes results more predictable.
“The result is a set of agents, each with a tiny, predefined mission. A decision tree architecture that prevents hallucinations and offers observability and predictable results, essential for sensitive infrastructure.”
This approach addresses three major challenges for the industry. Firstly, observability: every action can be understood, tracked and audited. Secondly, resistance to hallucinations: complex tasks are broken down into simple, controllable and verifiable steps. Finally, traceability: every decision, every action and every piece of data used can be logged without adding to the operational burden on teams.
Beyond automation, agent-based AI thus becomes a decision-support tool. It documents, contextualises and structures useful information, particularly in situations where operational pressure is high. Humans remain at the end of the chain to monitor, arbitrate and validate, but they have a clearer, more reliable and more actionable framework.
As Marc Allaire points out:
“AI helps prepare the groundwork, but humans remain essential for monitoring and validating actions at the end of the process.”
Read the full article in Les Échos.
