In partnership with
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic say AI now writes 100% of their code.
Are you using AI to write yours?
These 100+ Claude Code hacks show you exactly how. Sign up for The Code and get:
100+ Claude Code hacks — free
The Code newsletter — learn the latest AI tools and skills to code faster in 5 mins a day
Google showed an AI that builds interactive worlds you can talk to and edit on the fly, Anthropic cut a deep discount to put Claude across the US government, and OpenAI quietly started leaning on its rival's chips. The world-model demo is the eye-catcher, but the compute and government moves may matter more. We checked the releases and the reporting before writing.
TODAY IN AI
3 things that happened while you were busy
1. Google DeepMind's Genie 4 builds interactive worlds with sound and speaking characters.
Genie 4 generates photorealistic worlds in real time that you can change mid-scene with a text prompt, altering the weather, moving the camera or adding a person. DeepMind says it is the first real-time world model to generate audio natively, including music, ambient noise and dialogue, and that scenes now stay consistent for up to two minutes, double Genie 3. It is a research preview for trusted testers, with Google AI Ultra subscribers ($100 a month) next through Project Genie. Read TNW's report. DeepMind admits details and object permanence still drift in longer scenes, and every clip carries a SynthID watermark.
2. Anthropic cut a 40% discount to spread Claude across the US government.
Anthropic widened its GSA agreement to cover federal, state and local government, offering 40% off Claude Enterprise, Claude for Government and the API, per The Neuron. It estimates about $1 billion in savings, and says roughly 500 US government organizations now use Claude. See The Neuron's roundup. The savings and adoption figures are Anthropic's own, and the discount deepens an earlier $1-per-user deal.
3. OpenAI reportedly turned to Google's chips to run its products.
OpenAI has started using Google's TPUs, its first significant use of non-Nvidia chips, according to reporting on The Information, as it looks to cut inference costs and lean less on Microsoft. Neither company has confirmed terms or volumes. Read DatacenterDynamics' report. Two fierce rivals sharing hardware says a lot about how tight compute has become.
TODAY'S TUTORIAL
Storyboard a video with AI before you shoot a frame
World models like Genie 4 point at where video is going, but you can plan a shoot with the tools you already have today. Here is a five-step way to turn an idea into a shot-by-shot storyboard in about 20 minutes. See where world models are heading.
1. Write the brief. In ChatGPT or Claude, describe the video in three lines: who it is for, the one message, and the length. Ask for a shot list, not prose.
2. Get the shot list. Ask for six to 10 shots, each with the camera angle, what is on screen, and one line of voiceover or text. Tell it to keep each shot to one simple idea.
3. Draw each frame. Paste each shot description into an image model such as Canva Craft and generate a rough frame. You are after composition, not a finished image.
4. Lay it out. Drop the frames in order into a slide or doc with the voiceover under each, so the whole video reads top to bottom on one page.
5. Pressure-test it. Ask the model to find the weakest shot and the point where a viewer would drop off, then fix those two before you film or hand it to an editor.
Good to know: Keep the frames rough. The storyboard is there to settle the order and the message cheaply, so you do not waste a shoot discovering the script does not work.
FROM THE FRONTIER
The flashy demos run on a compute crunch nobody has solved
The demo. Genie 4 is the kind of thing that makes headlines: talk to a world, change the weather, train a robot in simulation. DeepMind even had it simulate Apptronik's Apollo humanoid playing checkers as a test of large-scale robot training. See what Genie 4 does.
The bill. All of it needs chips, and chips are scarce. OpenAI turning to Google's TPUs, after years of relying on Nvidia and Microsoft, is a sign of how hard inference costs are to control. Read the TPU report.
The scale. Anthropic says power delays and grid costs could slow its own buildout, against chip and data-center commitments near $400 billion, and The Washington Post estimates the industry could seek roughly $10 trillion for chips and data centers through 2032, per The Neuron. See the figures.
The customers. The way to pay for that is volume, which is why Anthropic is discounting Claude 40% to put it across government, and the Dutch government is planning to cut 25,000 state jobs it links to AI savings. The business model is scale on both ends. See the government moves.
The reframe. It is easy to watch the world-model demo and think the hard part is the AI. The hard part is the power plant, the chip supply and the customer big enough to pay for both. The magic is cheap now; the electricity is not.
MONEY FORWARD
Three ways to turn this week's AI news into income
1. AI storyboards for video creators. Small businesses and creators waste money shooting videos that do not land. Sell the tutorial above as a service: hand them a shot-by-shot storyboard with rough frames and voiceover before they film. See where video AI is going.
2. Cheaper automations on halved prices. OpenAI's GPT-6.1 Terra costs half of the last version, so routine automations get cheaper to run. Rebuild a client's workflow on it and keep the margin, or lower your price to win the job. See the cost pressure driving this.
3. Design packs with Canva Craft. Canva's image model renders legible text and is free to all Canva users, so you can sell local shops a monthly bundle of flyers, menus and social posts without stock-photo fees. Read about Canva Craft.
GET THE EARNINGS GUIDE
Want the step-by-step version of ideas like these? Get The Aiwire earnings guide at theaiwire.co. These are ideas, not income promises, and results depend on your skills, market and effort.
Discover laptops with built-in intelligence.
Shift your workforce into hyper-productivity mode with Dell Pro laptops powered by Intel® Core™ Ultra with Intel vPro®. Shop on Dell Premier.
IN THE KNOW
What people are actually watching and sharing
Call it "Super Intelligence." Trump said the government will stop using "AI" in official documents and prefers "Super Intelligence," per The Neuron, though no implementation detail was shared. See the roundup.
A 15,000-page proof, machine-checked. Axiom says it computer-verified the Classification of Finite Simple Groups, whose original proof runs about 15,000 pages, as a Lean formalization of roughly 43 million lines built with AI agents over 24 days. Read about the effort.
Cloudflare's crossed wires. A config issue plus a software bug let some long AI requests on Cloudflare receive other customers' cached data. Cloudflare found 155 affected responses over about two weeks and no sign of deliberate abuse, per The Neuron. See the details.
AI on a phone. Liquid AI released LFM3, a family of small models for text, vision, audio and more, sized to run on phones and laptops; firms above $10 million in revenue need a commercial license, per The Neuron. See Liquid's models.
25,000 jobs. The Dutch government plans to cut about 25,000 state jobs, over 15% of its civil service, as part of AI-related savings, per The Neuron. Read more.
TRENDING TOOLS
5 tools worth a look this week
1. Genie 4. Google DeepMind's world model generates editable, talking worlds in real time with native audio. It is a research preview for trusted testers, with Google AI Ultra ($100 a month) next, so this one is to watch. Read the details.
2. GPT-6.1 Terra. OpenAI's new mid-size model for visual reasoning and computer use, 50% cheaper than the last Terra and available in ChatGPT Work, Codex and the API. The accuracy gains are OpenAI's own figures. Read more.
3. Kimi K3.1 Pro. Moonshot's near-frontier model, strong at agentic coding, at $2 per million input tokens and $10 output. Free weights are promised by November 4; the benchmarks are the company's own. See Kimi.
4. Canva Craft. Canva's first in-house image model plans the layout before drawing, so text comes out legible, and it is free to all Canva users with a public API. MCP access for agents is coming. Read the overview.
5. Liquid AI LFM3. A family of small models for text, vision and audio built to run on your phone or laptop rather than the cloud, useful when privacy or offline use matters. A commercial license is needed above $10 million in revenue. See the models.
PROMPT STATION
Turn one idea into a full video shot list
This prompt works in Claude or ChatGPT and turns a rough concept into a shot-by-shot plan you can film or hand to an editor. Use it with the tutorial above before your next video.
COPY AND PASTE THIS PROMPT
You are a video director. I want to make a [LENGTH, e.g. 30-second] video about [TOPIC] for [AUDIENCE], with one main message: [MESSAGE]. Give me a shot list of 6 to 10 shots. For each shot include: the camera angle or framing, what is on screen, one line of voiceover or on-screen text, and roughly how many seconds it runs. Keep each shot to a single idea, make the first three seconds grab attention, and end with a clear call to action. After the list, name the one shot most likely to lose the viewer and suggest a fix. Keep the whole thing practical to film with a phone.
Tip: Swap [TOPIC] and [MESSAGE] for a product demo, a tutorial or a founder story, and set [LENGTH] to match the platform, such as 15 seconds for Reels. For a sharper plan, add [the action I want viewers to take] so the call to action is baked into every shot.