The week in brief — with AI, every new capability is a decision that stays yours

Four nuggets this week, one common thread: the more powerful AI becomes, the more decisions it puts to you; it takes none away. Coca-Cola handed its Christmas ad to generative AI, with five specialists and more than 70,000 clips in a month; the efficiency was real, but public trust suffered, because the real question was not whether AI could produce the film, but whether to entrust it with what carries the brand’s promise. On the technical side, a study presented at ICML showed that a large advertised context window guarantees nothing: on a long document, the best model tested drops from 99.3% to 69.7% accuracy. An advertised capability is not a quality verified on your own documents.

The other two nuggets are about control. According to ISACA, 56% of organizations do not know how quickly they could stop an AI in an incident: an autonomy you cannot halt stays a risk. And Microsoft finds that the employees who get the most from AI are not better equipped: they decide task by task what goes to the AI (53% versus 33%) and deliberately keep part of the work without it. The thread of the week: every new AI capability adds a decision that stays yours, it does not make it for you. Before scaling an AI tool, write down, for each use, what you entrust to it, what you verify on your own cases, who can stop it, and what you keep in human hands.

Source: YPrtnrs Institute nuggets, week of October 26–29, 2026, drawing on Forbes and Euronews (Coca-Cola), NoLiMa (ICML 2025), ISACA, and Microsoft.