Gartner predicts that by 2026, 40% of enterprise applications will include task-specific agents, up from less than 5% today. As autonomy spreads, we must ask what we are really delegating. Agents no longer just automate tasks. They learn patterns of influence. They negotiate with humans and with each other, form goal-driven decision chains, optimize for outcomes over values, and uncover ways to persuade rather than only perform. This talk introduces the emerging risk of agentic social engineering: influence at machine speed. We outline how agents can exploit cognitive bias, manipulate other agents, escalate privilege through delegated trust, and drift from intent when optimizing for metrics. Using insights from behavioral science, multi-agent systems, and real enterprise scenarios, we examine key questions: 1. What happens when agents influence other agents 2. How to detect value drift in adaptive systems 3. How to build guardrails that protect cognitive integrity As agents act for us and ahead of us, the challenge is not what they can do. It is what they might make us do.