From firefighting to building: How AI agents restored our team’s core productivity

Grab's Analytics Data Warehouse team was spending ~40% of engineering time on repetitive support tasks. They built a multi-agent AI system using LangGraph and FastAPI that routes Slack requests through two pathways: an Enhancement Agent for code changes (creating GitLab MRs, running tests) and an Investigation pathway with four specialized agents (Classifier, Data Agent, Code Search Agent, On-call Agent, Summarizer). Key engineering challenges solved include context/token management, tool proliferation (trimmed from 30+ tools), SQL safety guardrails with PII protection, and a human-in-the-loop review model. The system reduced issue resolution from hours to minutes, reclaimed several FTE-equivalents of engineering bandwidth, and continuously improves via annotated feedback loops.