Grab's engineering team shares how they've built a five-level autonomy ladder for AI-driven analytics, moving from AI-assisted work (L2) to fully autonomous loops (L5). Key internal systems include Spartan, an agentic Slack-based analytics workflow with 50+ skills and 120+ analysis frameworks; Scarlet, a near-self-healing pipeline tool; ContextIQ for managing metric context lifecycles; and BriX, an internal portal that grew 10x since September 2025. Concrete results include mechanical ticket share dropping from 44% to 30%, cycle times reduced by ~33%, and self-serve agent resolution of SQL requests rising from 50% to 81%. The post argues that as agents own more of the analytics loop, analysts shift from producing artifacts to owning problem framing, judgment, and decisions.
Table of contents
IntroductionThe ladderMaking the climbWhat this looks like in practiceThe path forwardJoin us5.2K Impressions1 Comment