[{"data":1,"prerenderedAt":55},["ShallowReactive",2],{"insight-ai-operations":3},{"id":4,"title":5,"body":6,"category":45,"date":46,"description":47,"extension":48,"meta":49,"navigation":50,"path":51,"seo":52,"stem":53,"__hash__":54},"insights\u002Finsights\u002Fai-operations.md","AI becomes useful when it enters an actual operating model",{"type":7,"value":8,"toc":41},"minimark",[9,13,16,38],[10,11,12],"p",{},"The easiest AI prototype is often the least interesting production system.",[10,14,15],{},"A useful operational implementation needs answers to ordinary architecture questions:",[17,18,19,23,26,29,32,35],"ul",{},[20,21,22],"li",{},"What data may the model see?",[20,24,25],{},"Which actions may it perform?",[20,27,28],{},"Which actions require approval?",[20,30,31],{},"What happens when a tool fails?",[20,33,34],{},"How is a decision reconstructed later?",[20,36,37],{},"Where does deterministic software remain the better choice?",[10,39,40],{},"The model is only one component. The operating model around it is what makes the capability trustworthy.",{"title":42,"searchDepth":43,"depth":43,"links":44},"",2,[],"Applied AI","2026-08-10","Connecting intelligent tooling to business systems requires more architecture, not less.","md",{},true,"\u002Finsights\u002Fai-operations",{"title":5,"description":47},"insights\u002Fai-operations","hOkgiJMshOn-1zHlv8wmElJB07j4E9rr0U6EzXPWCsI",1787213493756]