Why Enterprise AI Fails: Stop Bolting AI On and Redesign Your Workflows

Most enterprise AI deployments fail — not because the technology isn't ready, but because companies are solving the wrong problem. Adding AI to broken workflows doesn't fix them. It just makes the breakdowns faster. Michael Krantz, Editor in Chief at Box, sits down with Nirmal Ganesh, Senior Director of Product Management, to unpack what's actually going wrong — and what it takes to get AI working at enterprise scale. Nirmal introduces two frameworks that explain why so many deployments stall: → The supervised autonomy trap — when you add AI to every step of an existing process and end up with more approvals, not fewer. It's a manual workflow with extra steps. → Trust engineering — the governance discipline that lets autonomous agents act at scale: defined permissions, audit mechanisms, confidence scoring, and escalation protocols baked into the architecture — not bolted on at the end. The discussion covers: → Why 77% of AI deployment failures are organizational, not technical → The difference between procedural and structural guardrails → How redesigning a 10-step process gives you a 4-step process — not a faster 10-step one → Why change management determines whether employees become AI advocates or resistors → How Box Automate inherits existing permissions so agents can only act on what users can access → What it takes to move from an isolated AI pilot to enterprise-wide deployment If you're leading AI transformation, operations, or workflow automation, this is the conversation that reframes the question — from "how do I make AI faster?" to "what processes do I actually need?" 🔗 Learn more about Box Automate: box.com/automate