a16z heads a $20M seed funding round for the AI agent startup Runta.

      AI agents can now handle travel bookings, write code, and spend money on your behalf. While this capability is powerful, it also poses risks. A new startup called Runta has secured $20 million in funding to mitigate potential damage caused by these agents. According to The Information, the investment led by Andreessen Horowitz (a16z) has valued Runta at over $100 million.

      'Parenting' AI Agents

      Guanlan Dai, the founder of Runta and a father of two, likens AI agents to very precocious children. Just as parents make their homes safe and keep credit cards out of reach, he believes developers should restrict the files an agent can access and limit its spending capability. His company assists businesses in "parenting" their agents by placing them in isolated environments with protective measures, ensuring that a rogue agent cannot incur large expenses or disrupt systems.

      A Computer for Every Agent

      Runta aims to reconstruct the underlying layer where agents operate. In a16z's announcement of the funding, partner Martin Casado mentioned that agents "just want a computer." This refers to a complete operating system, stateful, capable of functioning locally or in the cloud, with integrated security controls. Casado emphasizes that this is not merely another "sandbox cloud," but rather a foundational systems software designed specifically for agents. He also notes an unexpected consequence of the growing demand for agents: a CPU shortage is emerging alongside the GPU shortage driven by models, as agents require substantial conventional computing power.

      Dai is well-acquainted with this layer. He previously served as a technical lead on Cloudflare’s edge team and contributed to building the core proxy at the API company Kong.

      The Enterprise Agent Land Grab

      Runta is part of an expanding field of startups focused on providing the necessary infrastructure to safely operate AI agents within organizations. Others are addressing agent authorization, access governance, and oversight of an AI workforce. The message is consistent across these companies: organizations are assigning genuine tasks to autonomous software and want to maintain control while minimizing potential harm when agents deviate from expected behavior. Casado describes the transition from hosting software to hosting agents as the most significant shift in computing to date.