Every Enterprise Is About to Have a 100,000 Agent Problem | Oren Michaels of Barndoor AI

AI agents can now connect to every enterprise tool your employees use. The problem is nobody built the governance layer that tells them what they're actually allowed to do. In this episode, Craig Smith sits down with Oren Michaels, co-founder and CEO of Barndoor AI, to explore what he calls the 100,000 agent problem: the inevitable reality that as enterprises deploy AI agents to act on behalf of employees across tools like Salesforce, Slack, and email, the number of agents requiring their own rules and guardrails scales to a size no human team can manage manually. Michaels explains why MCP (Model Context Protocol) solved connectivity but did nothing for trust, why traditional identity and access management systems were never designed for the failure modes that AI agents create, and why the result is a fundamentally new security problem: authorized people using allowed tools in ways that still cause damage. He also introduces a consumer-facing product called Venn that lets individuals experience what governed agentic AI actually feels like, so they can bring that understanding back to their organizations. Key Topics Covered: ● The 100,000 agent problem: why every enterprise deploying agents at scale will face a governance crisis they can't manage manually ● Why MCP solved connectivity but not trust, and why that gap is the most important unsolved problem in enterprise AI ● Why traditional identity and access management fails with AI agents: humans come with built-in judgment; agents don't ● The new enterprise security vector nobody prepared for: authorized agents doing authorized things in unauthorized ways ● Why 95% of AI projects fail, and what it actually takes to make adoption succeed across an organization ● Why 2026 is the inflection point: knowledge workers are finally seeing agentic AI solve their own problems, and the adoption curve is about to go vertical Enterprises want agents but won't deploy them without trust, and trust requires governance that most current infrastructure was never designed to provide. Barndoor AI is building the specific layer that closes that gap, and this episode is one of the clearest explanations available of why that layer is the critical missing piece. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI. Craig Smith on X: https://x.com/craigss EYE On A.I. on X: https://x.com/EyeOn_AI Timestamps 00:00 Why AI agents still struggle in enterprises 02:22 The “100,000 agent” problem explained 03:17 Why trust is the biggest issue in AI adoption 05:35 How Barndoor governs AI agents safely 06:40 What Venn does for individual users 08:12 Why enterprises hesitate to deploy AI agents 09:32 AI agents as “enthusiastic interns” 12:12 How people slowly build trust in AI 13:09 Why identity systems aren’t enough for AI 15:20 Why AI needs a smaller “blast radius” than humans 16:21 How companies can monitor AI mistakes 18:41 How Barndoor controls agent permissions 20:24 Safe read actions vs dangerous write actions 25:42 What causes context window exhaustion 27:48 How Tool IQ simplifies AI tool usage 29:00 Why OpenClaw changed the AI conversation 31:17 Why enterprises are finally embracing AI agents 37:23 Why 2026 could be the tipping point for AI adoption 41:07 The future of agentic workflows in enterprises 45:06 How hotels could use AI for hyper-personalized service 47:14 Who will manage AI agents inside companies 54:08 How AI is already transforming Broadway productions 59:05 The meaning behind the name Barndoor AI