Before the call ends, Claude has already drafted the action plan. It joins as a visible participant, posts a notice in the meeting chat, and can answer questions using your company's connected apps — think pulling last year's figures while someone asks about them mid-call. After the meeting, the notes, action items, and any artefacts wait for you at claude.ai. Why you care: the most expensive part of most meetings is what doesn't happen in the hour after them — nobody writes the plan, nobody assigns the tasks. Claude does it while you're still talking.

You add your card once. When you ask your Dot to buy something, it finds the item, shows you the total, and waits for you to approve it with a security check before paying. Your real card details stay private — the card networks hand your agent a one-time payment token, scoped to that single transaction, instead. Amex, Mastercard, and Visa are the launch partners, and OpenAI is taking no cut of transactions for now. Why you care: this is the first time a mainstream AI assistant has a real payment method attached to it. Agents that can find, buy, and confirm in one move are here — this is the opening chapter.

HomeView is a 7-inch screen for your kitchen, hallway, or bedside that runs Siri AI. It went on sale yesterday in the US, UK, Canada, Australia, India, New Zealand, and Singapore. Inside is the same A18 chip as the iPhone 16 — real on-device power, not a low-end gadget chip. It handles smart home controls, reminders, calls, and questions, and Siri AI can pull personal context from your nearby iPhone when you need it. Why you care: Amazon and Google have owned the smart display market for a decade. Apple has never played. This is the opening move — and it's priced to compete from day one.

Hydrogen is Google's new workhorse: fast enough for everyday work, capable enough for coding and very long documents. It has a 1-million-token context window and four thinking levels so you can trade speed for depth. For developers it costs $1.50 per million input tokens and $7.50 per million output tokens — a flat rate across the full context window. For everyone else it's already baked into Google Search for free. Google says it scored 57% on Terminal-Bench, its test of agent performance on terminal tasks, versus 18% for the Gemini 3.8 Flash model it replaces — roughly three times better. Why you care: the fast-and-cheap tier of AI just got meaningfully more capable. If you use any Google product, it's already running under the hood.

We vibe-coded a game in the open — Nova's World — a little browser RPG we built by describing it in plain words, no hand-written code, growing it in public. Walk the forest, fight bosses, climb the leaderboard. play it free →

Open Google AI Studio (aistudio.google.com) and select Gemini 4 Hydrogen — it's free to test. Paste in the longest document or codebase you've been trying to fit in context and ask it to summarise, find gaps, and list action items. The 1-million-token window handles things other models have to chunk.