There is a less visible part of the electricity grid: the one that goes out on purpose. From time to time, whole teams must disconnect by maintenance. They are routine tasks, but their coordination and validation are decisive so that the system continues to work. In California, that work falls to the California Independent System Operator (Caiso) human team. Now, an artificial intelligence will begin to collaborate with them.
Caiso has launched a pilot project with the Oati firm. The objective is to verify how a language model, similar to the one that drives chatgpt, can attend a critical mission: manage the requests – especially the programmed – that come daily to perform maintenance on the network.
Artificial intelligence to attend a task that cannot fail
Many of these requests imply, although not always, temporarily disconnect key assets. If the information is incomplete or incorrect, the consequences can be felt in real time: from Network instability even alterations in the electricity market. Therefore, although part of the flow is already automated, the final review remains manual and exhaustive.
Between 2020 and 2025, internal graphs show a high volume of applications, with clear spring and autumn peaks, the usual maintenance windows. That seasonality, added to the total number of requests, converts maintenance management into one of the most complex operating challenges in the system.
The tool is called Oati Genie and is raised as a Copiloto of Ia. Detect anomalies, extract operational information and suggest steps based on previous cases. To achieve this, it combines retrieval-augmented generation (RAG) with a system of multiple agents: operators formulate questions in natural language and receive contextualized answers, with links to documents and records that support them.
The operator can, for example, ask what work could affect a certain electric route and obtain a history, procedures and public data such as Oasis (Open Access Same-Time Information System), the regulated base that centralizes technical and market information in real time. That consultation, which previously demanded manual searches, now comes with much more agility. The co -pilot does not decide for itself, its role serves to reinforce and expedite the human decision.
If everything is going as planned, the tool will enter internal tests before the end of the year. Those who use it may report failures, limitations or improvements and help profile their evolution. It is not an isolated experiment, but part of a broader strategy to Apply to other areas of the system.
The plan is underway, but there is still knowing if Oati Genie will fulfill what he promises. It may become a daily ally for operators; Maybe it’s just a step on the road to a smarter network. The only sure thing is that the electricity sector already explores that address.
Images | Caiso | Javad Esmaeili
In WorldOfSoftware | Harvest wheat or kilowatts? The new account that many farmers make in Spain
