Palo Alto Networks launches Cortex XCOR with Martin Mao, claims 75% root-cause accuracy

Martin Mao spent more than a decade working on observability, from internal monitoring for EC2 at AWS to the observability platform for Uber, and co-founded Chronosphere with Rob Skillington to manage cloud-scale telemetry costs. That cost-management foundation remains a pillar of XCOR, which the company says optimises customer data by an average of 89% without sacrificing visibility, and whose customers include OpenAI, DoorDash and Compass. Mao argues the turning point came in late 2025 with frontier models such as OpenAI's GPT-5 and Anthropic's Claude Sonnet and Opus, whose reasoning capabilities advanced enough for agents to act as autonomous experts rather than assistants.

The launch is also a response to a surge in AI-generated code. Mao cites research showing 42% of developers now report that AI writes at least half their code, up from 12% last year, and argues that rapid releases expand the surface area for production failures and increase the burden on SREs. XCOR is designed to match that velocity by combining the XCOR Operator, specialised agents, and the XCOR Fabric, which grounds agents in a Knowledge Graph, Operational Memory, User Behavior and Human Knowledge.

The stakes hinge on whether the reported performance holds up outside the environments where it was measured. The 75% success rate and under-three-minute average are the figures the platform will be judged against, and the further 19% of incidents flagged as useful but not successful suggests a meaningful share of investigations still require human judgement. For operations teams evaluating the platform, the deciding factor is likely to be whether automated root-cause analysis reliably reduces the manual triage that consumes senior engineers' time.