Agentic AI is easy to demo and hard to run at scale in a governed environment. This session is about running it. At MoneyHero Group, AI now writes most of our new code and answers most first-line customer questions across four markets. Those gains only arrive after you solve the harder problems of architecture and ownership. When agents do the building, the line between product and engineering stops holding. A product spec becomes the thing that ships. Code review turns into reviewing intent. A loose requirement fails in production instead of in a planning meeting. Artem covers the decisions behind that change: where humans stayed in the loop, where they came out, what had to be rebuilt, and the team changes it forced. The focus is operational, not speculative. What broke, why it broke, and what it took to make agentic AI reliable in production.