Your AI Agents Are Bleeding Tokens: Building a Runtime Cost Control Plane for LLMs

As teams ship more autonomous AI agents, a new failure mode has emerged: silent, expensive runaway spend — infinite loops, exploding context windows, and redundant system prompts that quietly triple your LLM bill. Most AI observability tools show you what happened after the fact. In this talk, I'll walk through building Aurex, a runtime cost control plane that intercepts and prevents cost runaways before they hit your invoice — enforcing budget policy at the SDK level, detecting context bloat and redundant prompts, and giving engineering leaders a real "cost-per-successful-outcome" view instead of just raw token counts. I'll run a live audit that catches a runaway agent loop in under a minute, share what we learned building local-first, privacy-preserving cost governance for AI-native teams, and discuss where this category is heading as agentic AI budgets become a board-level conversation.