In this episode of The Agentic Review podcast, Zichuan Xiong, Global Head of AIOps at Thoughtworks, joins Itamar Friedman and Nnenna Ndukwe to look at what AI is really doing to the cost of running software. Over two years, Zichuan’s team grew from a single RAG use case to 25 to 35 operational agents for enterprise clients. He has seen both the job-level gains and the system-level bill: one customer raised AI adoption 1.5x in eight months while operations costs rose 10x. He makes the case for an operations-first SDLC and explains why semantic drift slips past LLM tracing. Zichuan also argues that turning tribal knowledge into agent-ready context needs a dedicated new role.