The source says Devin Oncall is generally available and can investigate alerts and bug reports, diagnose incidents, and help resolve them. Cognition reports 100% end-to-end success on SREGym, whose tests require identifying a root cause and mitigating the fault in the same run. It also reports that AngelList’s system identified root causes 3.5 times more often than its in-house harness and handled 50% of incidents end-to-end. These are concrete claims in the product announcement, not independent validation.
Implication: Teams should assess permitted actions and the review boundary, not only benchmark accuracy. The source establishes Cognition’s reported results and examples, but does not independently verify them or demonstrate performance across all incident types and production environments. Watch whether customers report sustained outcomes, including safe mitigation rates and how often human review or escalation is needed.
Thesis
Devin Oncall’s launch treats production incident response as a coding-agent workflow, where shared system context and review mechanisms matter alongside model capability.
Observed facts
Cognition announced Devin Oncall's general availability for investigating alerts and bug reports and helping resolve production incidents.
1Cognition reports 100% end-to-end success on SREGym, whose tests require identifying a root cause and mitigating the fault in the same run.
1Cognition reports AngelList's system found root causes 3.5 times more often than its in-house harness and handled 50% of incidents end-to-end.
1
Mechanism
Cognition describes an agent with relevant system access, shared team context, and production investigation workflows, with a second reviewer agent checking for reasoning errors and engineers reviewing analysis or fixes and receiving escalations.
Why now
Cognition says the product is generally available and handles thousands of real production alerts daily, making its operational claims timely for teams evaluating agent-based on-call work.
Implications
Teams should assess permitted actions and the review boundary, not only benchmark accuracy.
The shift places investigation and mitigation in an agent workflow while retaining human review and escalation.
Evidence boundary
The source establishes Cognition’s reported results and examples, but does not independently verify them or demonstrate performance across all incident types and production environments.
Watch conditions
Watch whether customers report sustained outcomes, including safe mitigation rates and how often human review or escalation is needed.