Sberbank has introduced AI-Disrupt PSLC, an operational model designed for managing the software lifecycle in the era of AI agents. The approach shifts from traditional IT Service Management by prioritizing an AI-first principle, where support processes are specifically engineered for collaboration between human operators and autonomous agents.
The framework focuses on four key areas: team structure, operational processes, platform architecture, and the integration of support with development. Engineers are responsible for creating knowledge bases for agents, designing execution scenarios, and validating performance. Support processes are transformed into executable runbooks that include predefined authorization levels and automated outcome verification.
According to Kirill Menshov, Senior Vice President and Head of Technology at Sberbank, this methodology impacts areas accounting for 30–40% of total IT costs. The model builds upon the bank's previous AI-Disrupt PDLC initiative, aiming to link the creation of new software features with their long-term reliability and maintenance after deployment.
The model mandates that humans define goals, success criteria, and operational boundaries while retaining final responsibility for results. AI agents operate within these constraints, with their level of autonomy determined by the risk and reversibility of specific tasks. Continuous monitoring and metrics are used to ensure system reliability and build trust in agent performance.