"AI in everything" sounds like a slogan, and used carelessly it is one. We mean something specific and boring: at each step of a system, we ask whether the behaviour should be hand-written or learned from data.
Often the answer is still "hand-written" - deterministic code is cheaper and more predictable for plenty of tasks. But asking the question consistently surfaces the places where a model removes real complexity: classification, extraction, ranking, summarization, anomaly detection.
Applied this way, AI stops being a product feature and becomes an engineering default. The interface might not even mention it. The system is just smaller, faster, and harder to break - which was the point all along.