FixItFlow uses LLMs to auto-generate cloud troubleshooting guides
Engineers cut incident mitigation time by 2.3x with AI-generated docs
FixItFlow, a new system from researchers at (affiliations not explicitly stated, but from the paper's authors), automates the creation of troubleshooting guides for cloud incidents using large language models (LLMs). The system ingests historical incident data—including the sequence of diagnostic actions taken by engineers—and extracts reliable patterns to build structured, step-by-step guides with verified commands. A key innovation is a strict validation layer to prevent LLM-generated hallucinations or fabricated content, ensuring only actionable, safe steps are included.
In an evaluation with 26 engineers, FixItFlow-generated guides achieved a 61.5% positive rating for clarity and demonstrated a 2.3x reduction in mean time to mitigation (MTTM) for incidents that had associated guides. The results suggest that automated guide generation can significantly improve cloud incident response consistency and speed while reducing the labor-intensive manual documentation process. The paper, submitted to arXiv on May 3, 2026, spans categories including Computation and Language, Artificial Intelligence, Machine Learning, and Software Engineering.
- FixItFlow uses LLMs to extract diagnostic patterns from historical cloud incident data and synthesize structured guides with verified commands.
- A validation layer prevents fabricated or hallucinated content, ensuring only safe, actionable steps are included in guides.
- In a study of 26 engineers, FixItFlow guides achieved 61.5% positive clarity ratings and reduced incident mitigation time by 2.3x.
Why It Matters
Automated guide generation could dramatically speed up cloud incident response and reduce engineering documentation overhead.