The way the full process works is that the raw data feed comes in, and machine learning is used to identify an anomaly that could be a possible incident. That’s where the generative AI agents step up. In addition to the history of similar issues, the agents also look for other relevant context information, such as other incidents on the network, research possible diagnoses, do root cause analysis, plan a remediation, calculate the confidence level of its recommendation, and explain the basis for that confidence number. And in this process, it’s not just one agent, but multiple agents checking each other’s work.
If the confidence level is high, the agent triggers an action. “We’ve done automation for a long time, and we have a library of actions,” says Abdelaziz.
If the confidence isn’t high enough, and if the action can have a big impact, it goes to a human being, where the generative AI has enriched the ticket fields. If the engineer agrees with the diagnosis and approves the recommendation, that decision is then fed back into the system for future learning.
Right now, this agentic system is only used for limited use cases, not the entire network. For extra security, the automated actions are scheduled during maintenance windows, so there’s no impact on customers.
“We’re doing it gradually,” Abdelaziz says. Over the past year, the agentic system has processed around 6,000 incidents. At the beginning, its success rate was around 88%, he says, and it’s now at more than 95%.
Next, the company is working on reducing its energy footprint by using agents to make energy decisions without jeopardizing network quality. And this is just the beginning. “I believe the best uses are yet to be discovered,” says Abdelaziz.
