How We Took an AI Agent From 41% to 97% Accurate - Milind Sunil Shah, Brainbase Labs

Milind Sunil Shah of Brainbase Labs walks through a six-step framework for improving back-office agents, using a Domino's project that tracks competitors' pizza prices across websites and delivery apps. From version 1 to version 11, accuracy went from 41% to 97%, cost per source halved from $2 to $1, and run time fell from 129 to 71 minutes, all through instructions, skills and playbooks rather than a new model. The steps: kickoff, ontology, benchmarks and evals, a narrow first agent, iteration, and handing off the whole loop. Chapters: 0:00 Intro 0:26 The Domino's pricing problem 1:36 From 41% to 97% 2:02 The six-step framework 2:23 1. Kickoff: agree on the win 3:35 2. Ontology: define every term 5:43 3. Benchmarks and evals 7:42 4. Ship a narrow first agent 8:49 5. Iterate: score, fix, ship 9:44 Fix: the deal below the link 10:28 Fix: stuck zooming into images 11:56 6. Hand off the whole loop Milind Sunil Shah Join us in SF for the second annual AIE CODE summit: https://aie.codes our last flagship summit of 2026. Applications live now!