The emerging risks from AI agents extend far beyond doomsday scenarios, as demonstrated by recent security incidents at OpenAI and highlighted in new research on agent autonomy. Rather than requiring an existential-level threat, autonomous agents operating within their intended parameters could fundamentally disrupt business systems built around human friction and approval workflows. The core vulnerability isn't malice or misalignment at scale—it's the friction-free execution of agent actions in environments designed with human checkpoints as safety mechanisms. Recent security incidents at OpenAI underscore practical vulnerabilities in agent deployment, particularly around permission boundaries and operational controls. As Google and Microsoft unveil new agent capabilities, the industry faces a critical question: how much friction can be removed before safeguards collapse? The permission systems currently limiting agent scope remain nascent and untested at scale. The geopolitical dimension adds urgency, with US-China AI talks ongoing as the two powers race to deploy agent technology. A Trump-Amodei meeting signals growing executive awareness of AI agent implications beyond technical circles, reflecting concerns that agent adoption could reshape competitive advantage in the AI era.