Maxim Bar Kogan, founder of Onyx Security, discusses the growing risks from AI agents in enterprises, such as accidental data leaks or unauthorized code publishing. He explains that existing security tools lack the context to understand agent intent, making them inadequate. Onyx addresses this by training specialized models to oversee other agents, acting as an AI control plane that balances performance, cost, and latency. The discussion covers the rapid adoption of autonomous coding agents, the need for vendor-independent oversight, and the importance of foundational security measures. Kogan argues that model labs are unlikely to solve trust and governance issues themselves, as enterprises require independent, cross-vendor solutions. He also highlights Israel's strong talent ecosystem for AI and security, and expresses a strong belief in the inevitability of AGI, urging companies to invest in foundational controls now.