Automated System Generates Cloud Incident Troubleshooting Guides in Seconds

Engineers managing cloud services face constant pressure to resolve incidents quickly, yet creating comprehensive troubleshooting documentation is a tedious, time-consuming task that falls behind as incidents evolve. A new research paper describes FixItFlow, a system that uses large language models to automatically generate incident response guides from historical data.

How It Works#

FixItFlow takes a different approach to the documentation bottleneck. Rather than relying on manual write-ups, the system analyzes patterns in how engineers have diagnosed and fixed past incidents. It then synthesizes this knowledge into structured troubleshooting guides, complete with validated commands and step-by-step procedures. A key safety mechanism prevents the system from inventing solutions—it enforces strict validation to ensure every piece of guidance is grounded in actual incident history rather than hallucinated fixes.

Results from Testing#

The researchers evaluated their approach with a group of 26 engineers working on real cloud infrastructure. The automatically generated guides received positive ratings for clarity from about 6 out of 10 evaluators. More significantly, engineers using the generated guides resolved related incidents 2.3 times faster than those without them—a substantial improvement in response speed.

Why This Matters#

Cloud operations teams today struggle with the gap between incident frequency and documentation capacity. Teams either leave gaps in their troubleshooting knowledge or spend engineering hours writing guides that quickly become outdated. FixItFlow suggests a path forward: by learning from incident data engineers already generate, automation can turn lived experience into shareable knowledge at scale. The results suggest this approach meaningfully accelerates how teams respond when problems occur.