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

Grab's Analytics Data Warehouse team describes how they deployed a multi-agent AI system to reclaim 40% of engineering time spent on repetitive data support tasks.

• •The system uses FastAPI and LangGraph to manage complex state and multi-agent collaboration, with Redis for caching and PostgreSQL for persistent memory.
• •Two pathways handle incoming Slack requests: an Enhancement pathway for code change requests and an Investigation pathway for data questions.
• •The Investigation pathway coordinates four specialized agents: Classifier, Data Agent, Code Search Agent, and On-call Agent, culminating in a Summarizer Agent.
• •Specialized agents were chosen over a monolithic AI for maintainability, modularity, and specialist accuracy despite added coordination complexity.
• •