Sergey Zhitinsky, founder of Git in Sky, shared insights from the OPS Talk by Sber meetup held on 19 October, where he examined the PSLC methodology presented by architect Alexander Shapovalov.
PSLC (Product Service Lifecycle) addresses the full service journey by combining immediate incident response with long-term improvements to processes, knowledge, and product design. It features two interconnected loops: one for service execution and another focused on building reliability and trust.
In contrast, Agent-Ops provides a granular eight-step framework for handling specific infrastructure tasks. These steps include defining intent, gathering evidence, performing diagnostics, creating a plan, obtaining approval, executing a controlled change, verifying outcomes, and extracting lessons. The methodology explicitly separates planning, human decision-making, and execution.
The article demonstrates their combined application through a realistic incident: after a release, increased logging fills the database disk, slowing order processing in an online store. PSLC manages the overall service recovery and subsequent reliability work, while Agent-Ops guides the infrastructure change such as expanding storage or safely removing old logs.
Both methodologies stress the importance of documented knowledge, controlled actions, and clear human responsibility. Agent-Ops v0.5.1 requires human approval and manual triggering of changes even in production, distinguishing it from fully autonomous execution models.
Resource considerations differ: PSLC benefits from centralized platforms and expertise pools typical in large organizations like Sber, whereas Agent-Ops aims to support compact teams that may lack such infrastructure.
