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

What It Covers

Oren Michaels, cofounder of Barndoor AI, explains why enterprises deploying AI agents face a governance crisis at scale. A 1,000-person company could generate 100,000 agents, each requiring distinct permissions. Barndoor provides a control layer between agents and enterprise tools, enabling safe autonomous action without rebuilding rules for every new AI model.

Key Questions Answered

• •The 100,000 Agent Problem: A 1,000-employee company where each worker runs roughly 100 task-specific agents produces 100,000 agents requiring individual governance policies. Each agent needs its own permission set defining exactly which tools it can read, write, or delete from — making centralized governance infrastructure a prerequisite, not an afterthought, for enterprise AI deployment.
• •Why Identity and Access Management Fails for Agents: Traditional IAM systems are intentionally over-provisioned because humans carry contextual judgment — they know not to delete Salesforce records even when technically permitted. Agents lack that judgment and can execute destructive actions at machine speed, requiring a finer-grained governance layer that restricts each agent to only the specific actions its assigned task requires.
• •Safe Read vs. Destructive Write Controls: Barndoor's governance model distinguishes between read-only and write/delete actions at the individual tool level. For example, a post-sales-call agent can log calls to Salesforce but is blocked from creating new contacts, preventing duplicate records when the agent fails to locate an existing entry. Enterprises configure these policies via toggles, JSON rules, or API calls.
• •Context Window Exhaustion and Tool IQ: Loading multiple MCPs simultaneously floods a model's token window with tool manuals before any task executes, degrading accuracy and wasting compute. Barndoor's Tool IQ layer intercepts agent requests, identifies the minimal relevant tool subset, and passes only those to the model — reducing token consumption and preventing agents from calling the wrong service entirely.
• •Venn.ai as an Enterprise On-Ramp: Barndoor's consumer product, Venn.ai, lets individuals connect personal tools like email, Slack, and calendar to a governed MCP layer for free. The strategy is deliberate: employees who experience agentic workflows personally develop the vocabulary and confidence to advocate for enterprise-wide deployment, shortening the sales cycle for Barndoor's corporate governance platform.