From Kubernetes to AI Agents: Sheng Liang on Obot

Sheng Liang co-founded Rancher Labs, the Kubernetes management platform SUSE acquired for $800 million. His latest company, Obot, turns that infrastructure background on the next problem: securing and governing AI agents inside the enterprise. In this conversation Sheng covers the step-function leap in what models can now do, the security and governance risks that come with agents pulling untrusted code off the web, and how Obot's gateway and registry address them. He also gets into where AI adoption really stands today, how the developer role is shifting from coder to lead, and why AI has turned open source into a pure meritocracy. Timestamps 00:00 "Software will basically be free" 00:18 18 months on: the pivot to Obot 01:35 The step function: function calling, MCP, skills 03:40 Shadow AI and untrusted code 06:30 Obot Gateway and Registry 08:30 MCP clients, servers, and skills explained 12:50 Dynamically generated, disposable skills 14:50 How reliable is it really? 17:30 Security vs governance 23:30 When agents go rogue 25:10 The changing role of the developer 30:30 Adoption is slower than the hype 34:50 How Obot is deployed 36:30 Egress control and partners 39:20 The AI Infrastructure Foundation 42:30 Open source as a meritocracy 49:00 OpenClaw, Docker, and what's next