Comfey | Yuxuan's Homepage

When production incidents occur in large-scale cloud systems, on-call engineers face a flood of alerts, telemetry, and runbooks. Triaging the root cause quickly is critical but demands deep system knowledge that is hard to encode in static rules.

Comfey is an agentic framework that automates incident triage. It uses LLM-powered agents that dynamically query telemetry data, consult historical incident records, and follow structured reasoning steps to identify likely root causes and surface actionable next steps for engineers.

Key contributions:

An agent architecture that decomposes triage into sub-tasks (symptom analysis, hypothesis generation, evidence gathering)
Integration with live production telemetry and incident history at Microsoft scale
Evaluation against real incidents demonstrating improved triage accuracy and reduced time-to-mitigate

(Yao* et al., 2026)

Published at FSE ‘26 Industry Track.

References

2026

An Agentic Framework for Triaging Incidents in Production Cloud Infrastructure

Yuhan Yao^*, Yuxuan Jiang^*, Minghua Ma, and 6 more authors

In Proceedings of the ACM International Conference on the Foundations of Software Engineering, Industry Track, 2026

Bib

@inproceedings{Comfey2026FSE,
  author = {Yao, Yuhan and Jiang, Yuxuan and Ma, Minghua and Vaidya, Madhura and Deng, Jieren and Hu, Yigong and Bansal, Chetan and Li, Ze and Chintalapati, Murali},
  title = {An Agentic Framework for Triaging Incidents in Production Cloud Infrastructure},
  booktitle = {Proceedings of the ACM International Conference on the Foundations of Software Engineering, Industry Track},
  series = {FSE '26},
  year = {2026},
  bibtex_show = true,
}