Wix built AirBot, an AI-powered Slack agent that automates incident investigation for their 3,500+ Apache Airflow pipelines. The system uses LLMs with Model Context Protocol to analyze logs, identify root causes, and generate pull requests automatically. By leveraging Socket Mode for security, LangChain for reasoning, and tiered model selection (GPT-4o Mini for classification, Claude 4.5 Opus for complex analysis), AirBot saves 675 engineering hours monthly. The architecture includes custom MCP integrations with GitHub, Trino, Spark, and OpenMetadata, achieving a 15% fully automated fix rate and 66% positive feedback across 60 engineers.

8m read timeFrom wix.engineering
Post cover image
Table of contents
1. Introduction: The Challenge2. The Problem: Why Traditional Alerting Struggled at Scale3. The Solution: AirBot4. Architecture & Design: A Blueprint for Building Your Own5. Operational Workflows: A Day in the Life of AirBot6. Impact & ROI: By the Numbers7. Conclusion: A Blueprint for Scaling Ops

Sort: