From Coding to Running: Engineering Practices of AI-Native SRE Agents | AICon Shenzhen

As AI model capabilities become commoditized, the competitive edge is shifting from model power to building scalable intelligent systems and engineering. The 2026 AICon Shenzhen conference, scheduled for August 21-22, will focus on AI infrastructure, large model systems, agent engineering, data intelligence, multimodal technology, and industry applications. Tencent Cloud's technical lead for CloudQ, Yao Binbin, will deliver a talk titled 'From Coding to Running: Engineering Practices of AI-Native SRE Agent'. He will discuss how SRE Agents address the operational governance challenge after software goes live, unlike Coding Agents that improve development efficiency. His presentation will cover building a data foundation for the 'running world' by integrating infrastructure, observability data, and knowledge; creating a unified Running World Model using ontology, semantic layers, and panoramic graphs; supporting operations scenarios like architecture visualization, health inspections, root cause analysis, capacity planning, change risk control, and FinOps optimization; and using Harness for evaluation, replay, experience accumulation, and continuous evolution from 'can answer' to 'verifiable, reproducible, and sustainable'. A real-world failure case will illustrate the closed-loop process from a CPU alert to root cause judgment and disposal recommendations.