The approach combines development and support, provides for control over AI actions, and retains human responsibility for the result
Sber's team presented the first version of the AI-Disrupt PSLC operating model, designed to manage the software lifecycle using AI agents. The approach was discussed at the engineering day of the GigaConf conference.
Unlike traditional ITSM practices, the model is initially designed for collaborative work between specialists and AI. Humans define goals, constraints, and result criteria, while agents perform tasks within established limits. Their level of autonomy depends on risks and the ability to undo actions, with responsibility for the outcome remaining with the employee.
Changes affect team organization, processes, platform, and the interaction between development and support. Engineers prepare knowledge bases for agents, create scenarios for operations, and verify their results. Successful scenarios can be reused in other services.
The model also provides for the transfer of operational data to development. This allows for considering the causes of failures when preparing updates, and for incorporating diagnostic, reliability, and recovery requirements even before the product's first release.
According to Kirill Menshov, Senior Vice President, Head of the Technologies block at Sberbank, changes within PSLC affect an area that accounts for 30–40% of total IT costs. The new approach develops the AI-Disrupt PDLC model, used in software development, and links the creation of new capabilities with ensuring stable operation after launch. Trust in AI, as Menshov noted, should be based on metrics, control over system behavior, and the possibility of human intervention.
