Enterprises are beginning to hand more of their day-to-day cloud operations to artificial intelligence, moving towards what is described as a “human by exception” model, Anant Adya, EVP and service offering head at Infosys told TNIE.
In this approach, AI systems handle routine monitoring and fixes, while people step in only when decisions carry significant risk, he said.
The shift builds on existing hybrid and multi-cloud management systems focused on performance, scalability and reliability. “By integrating Agentic AI, robust and scalable self-managing capabilities are built to use large-scale telemetry with reasoning and execution agents to automate actions across complex cloud environments, supported by human oversight,” he said.
In simple terms, the systems collect large amounts of operational data from across a company’s cloud infrastructure, covering applications, networks and underlying systems, and use AI agents to analyse and act on that information. By unifying this data, enterprises create “a single, intelligent operating layer”, helping move operations towards a “human by exception” model.
The AI agents are designed to spot early warning signs, detect unusual behaviour and carry out rapid triage. Instead of waiting for engineers to manually review alerts, the system can identify patterns and take action in high-volume situations. These include proactive monitoring, fleet management, backup assurance, incident triaging, recovery handling and change analysis.
“They make operational decisions and execute in high volume situations like proactive monitoring (lead indicator identification), fleet management, backup assurance, incident triaging, recovery handling, and change analysis,” Adya said. He added that this “boosts system availability, provides operators with vital insights, and reduces fatigue, all without elevating enterprise risk.”
However, companies are not removing people from the process entirely. Adya said the importance of a decision determines whether humans remain directly involved. “The level of criticality and potential impact of a decision determines whether human judgment must remain involved,” he said. “When a decision affects business critical systems, security posture, data integrity, or has a large or irreversible impact, humans must stay in the loop.”
Examples include production database changes, broad network reconfigurations, disaster recovery failover and actions involving high value assets.
Alongside operational changes, the introduction of AI agents is affecting how cloud services are priced and managed. As more work is carried out by software agents rather than teams of people, commercial models are shifting. Adya said this evolution is moving “from unit-based to outcome- and agent-based pricing, resulting in significantly leaner, more efficient enterprise operations.”
Cloud cost management is also changing. Traditionally, financial oversight of cloud use focused on governance, with optimisation handled separately. “Introducing autonomous AI agents shifts this to an outcome-driven approach,” Adya said.
Agents can improve chargeback accuracy, especially for shared resources, and automatically carry out cost-saving measures such as rightsizing and allocating optimal compute resources. Their actions are monitored through optimisation dashboards, and guardrails such as budget limits, quota policies and approval workflows are used to prevent cost escalation.