Andrew Miles Davis covers a week of AI news that mixes useful new features with some uncomfortable warnings. He leads with ChatGPT's new Intelligent UI, which decides how best to answer and can show interactive results while it is still researching in the background. He also covers Google's Playground, which lets you describe a game in plain English and share or publish it, though for now it is limited to adults in the US. Andrew then covers the story of a California woman who spent five and a half hours on the phone with a scammer using a cloned version of her daughter's voice, sending $5,400 in four transfers before discovering her daughter was at work. He explains why he has agreed a safety question with his own family that only they would know. Other stories include an update to Meta's Muse that builds a page on every person in your life, including people who never signed up, Audioball's plan for interactive audiobook stories where you can talk to the characters, and an author whose book on AI and truth was found to contain 26 quotes that did not stand up after AI tidied his original wording. Subscribe to In AI Nutshell for the AI news that matters to marketers, every Friday in ten minutes.

Andrew Miles Davis uses a BBC story about a 21 year old graduate who spent £30,000 on a content creation degree to ask a bigger question about AI learning. If you spend hours mastering a tool that changes a month later, what exactly did you pay to learn? He argues that access to information and actual learning are not the same thing, and shows how skills turn into features, then buttons, then a single sentence, using infographics and elaborate prompt formulas as examples from his own 15 years of training. He introduces the disappearing tool test, which separates tool knowledge from transferable knowledge such as explaining problems, questioning answers, and knowing when your own judgement should override AI. He ties it back to his time at MySpace, where the platforms vanished but the problems of attention, understanding people, and selling survived. Subscribe to In AI Nutshell for daily ten-minute episodes on what is actually worth learning.

Andrew Miles Davis answers three questions from recent training sessions and keynotes. He starts with whether your next manager could be AI, concluding that it could handle functional work like allocating tasks, tracking performance, and analysing data, but that accountability is the real barrier, since you cannot discipline or hold an AI personally responsible when something goes wrong. He then explains how to make AI forget something you have told it, focusing on temporary chat modes found in ChatGPT, Claude, and Copilot, while warning that temporary does not necessarily mean no record exists, so confidential company information should never be treated as safe. He also mentions deleting saved memories in settings and asking the model not to remember as alternatives. The episode closes with the question he hears in almost every workshop, which single AI tool he would pick, and his answer that there is no single best tool, only the right tool for what you are trying to do. For him that means ChatGPT and Whisper Flow, for stylisation-heavy presentation work it would be Midjourney, and for video producers it would likely be Seedance. Subscribe to In AI Nutshell for daily ten-minute episodes built around the questions real people are asking about AI right now.