Using AI at Work: AI in the Workplace & Generative AI for Business Leaders 66: Using AI Agents at Work: Governance, Security, and Real-World Wins with Oren Michaels
46 snips
Aug 25, 2025 Oren Michaels, founder of Barndoor.ai and expert in AI governance, delves into the pivotal role of AI agents in the workplace. He explains how AI agents differ from tools like ChatGPT and stresses the necessity of governance alongside innovation. Oren warns about the risks of unsanctioned Model Context Protocol (MCP) usage, highlighting its impact on security. He shares insights on treating AI agents like eager interns, revealing real-world successes in sectors like finance and marketing. This conversation serves as a guide for executives aiming to embrace responsible AI integration.
AI Snips
Chapters
Transcript
Episode notes
Agents Act; Chat Interfaces Advise
- Agents act rather than only suggest actions, creating a larger potential blast radius than chat interfaces.
- Treat agents like enthusiastic interns who need context, supervision, and gradual authority increases.
Watch Agents’ Actions, Then Approve
- Monitor agent actions in real time and compare attempted actions to expected behaviors before allowing writes.
- Block or investigate attempts that fall outside expected context or role boundaries.
Map Roles, Then Narrow Agent Scope
- Map human role permissions into agent permissions but make the agent's scope a strict subset of the human's access.
- Further limit agent capabilities by task, system, and trust level to reduce risk.