Siddhartha Borah from Brackett: teaching agents work that was never written down (Heavybit DevGuild)

In this episode, Siddhartha Borah, cofounder and CTO of Brackett AI, joins us at Heavybit DevGuild. Siddhartha explains how Brackett is building agents that can learn complex business work — especially in areas like supply chain and operations, where the real context often lives in people’s heads rather than documentation. We talk about teaching agents by doing the work, turning messy human knowledge into reliable workflows, agent-in-the-loop systems, memory, knowledge graphs, and why the last mile of enterprise automation is so hard. Links: • Siddhartha Borah’s LinkedIn • Brackett AI • Heavybit DevGuild 00:00 Intro 00:30 What Brackett AI is building 01:40 Learning work that was never written down 03:30 Turning human context into repeatable workflows 05:20 Why real-world agents fail without hidden context 08:00 Teaching agents by doing the work 09:50 Handling edge cases and human handoff 11:30 Memory, knowledge graphs, and “déjà vu” for agents 14:00 Where to learn more