Wednesday, November 4, 2026
Five Stories • Five Minutes

Good morning, and happy Wednesday. Here are five AI stories worth knowing, short enough to finish before the kettle clicks off.

What happened. Google plans to switch off Gemini, Gemini Notebook, Gemini in Classroom and Google Search for K-12 student accounts next year, until a school administrator turns them on, TechCrunch reports. The catch: AI is now built into Search, so a school that wants no AI at all has to turn Search off too. Google is also bringing Guided Learning, a mode that coaches students step by step instead of handing over answers, to Search and Google Classroom.

Why it matters. Google's apps and Chromebooks run a large share of classrooms, so this puts the AI decision in each district's hands. But it isn't a clean choice. A teacher who doesn't want AI Overviews doing homework still wants students to search for sources. Google says it will share specifics with schools in the coming months so they have time to prepare.

What to do. If your child's school uses Google accounts, ask now which tools it plans to switch on, and how students will search if it says no to AI. If your kids already use Gemini at home, point them to Guided Learning, which asks questions rather than giving the answer.

What happened. Apple's Audio Intelligence features for the Apple Watch Series 12 and Ultra 4 are now in the iOS 27.3 developer beta, TechCrunch reports. The four opt-in tools spot important sounds like sirens and doorbells, identify songs, and listen to conversations: Live Rewind lets you catch up on what was just said, and Siri Recaps turns talks into notes.

Why it matters. A watch that follows the conversation around you raises the same worries as AI glasses. Apple says Siri Recaps leaves out sensitive details such as ID numbers and financial data, and the work runs on the watch, the iPhone and Apple's Private Cloud Compute.

What to do. If you own one of these watches, wait for the public release rather than a developer beta. Before you use the listening features at work, check whether everyone in a conversation has to agree to recording where you live.

What happened. Two team members used GPT-6 Astra to design a small device for talking to the Codex coding agent, as a side project, in under a week and without direct human edits, OpenAI says. Astra picked the parts, laid out the circuit board, designed the case and wrote the firmware. A contract manufacturer built prototypes in under a month.

Why it matters. Designing physical products, from CAD drawings to circuit boards, has been slower to get AI help than writing software. It wasn't flawless: The Decoder notes several revision rounds were needed because Astra missed some manufacturing limits and tolerances.

What to do. If you make physical products, try giving ChatGPT a part's requirements and asking for a first-pass spec and parts list. Then have someone check it against what your manufacturer can actually build.

What happened. OpenAI has hired Dan Driscoll, the 26th Secretary of the Army, as its senior counselor for national security modernization, The Next Web reports. He will advise on how AI can strengthen national and cyber security, with a focus on helping US allies use it responsibly. Driscoll left the Pentagon in September after months of tension with Defense Secretary Pete Hegseth, according to The Wall Street Journal.

Why it matters. Governments are becoming major AI customers. The hire builds on Daybreak, OpenAI's cyber resilience program with allies, which has opened to Ukraine and been pitched to US utilities.

What to do. Nothing to act on this week. If you work with utilities or public agencies, watch for Daybreak-style AI security offers reaching them.

What happened. Revolut made self-hosted models its default for AI coding instead of tools like Claude Code and Codex, and cut its coding inference costs by roughly 90 percent in under a year, The Register reports. It serves the open model GLM-5.3 to about 2,000 engineers on 56 Nvidia B200 chips.

Why it matters. Ops lead David Abitbol said some engineers had been spending $1,000 to $10,000 a month with little return, so tokens were rationed. Once they were cheap, people used AI freely for debugging and routine tickets. He also said a 256,000-token context window is usually all you need, even when a model offers a million.

What to do. If your team pays for AI coding tools, pull last month's spend and list what it bought you. You probably won't buy 56 chips, but the budget lesson travels.

Using ChatGPT or Claude is like taking a taxi: you pay for every ride, and someone else owns the car. Self-hosting is buying your own car. You download an open model, like the GLM model Revolut uses, and run it on computers you own or rent. It costs more up front and you handle the upkeep, but each trip gets far cheaper, and your data stays in your own garage. That's the trade Revolut made.

What's one AI rule you'd want your kid's school, or your workplace, to set? Hit reply and tell us.

Know someone who'd like a five-minute AI briefing? Forward this email. New here? Subscribe at takefiveai.com.