THE AI BRIEFING #2 · JULY 30, 2026
|
|
| |
| |
|
Hi,
|
|
Edition 2.
|
|
Two weeks ago I wrote that Europe has a problem because it's lagging behind on AI models and AI compute. Now I think the opposite … I think!
|
| |
|
| |
|
China doesn't need to earn any of it back
|
|
Something I've caught myself doing a few times lately: thinking I've got the world figured out, only to find two weeks later it could just as easily be the complete opposite. Last edition was about Fable 5, and about how America treated it as weapons technology before the rest of the world was let back in. America simply switched the model off, the moment it suited the Trump administration. This time it's about China's countermove in a crazy macro-economic chess game, and where I typed Europe as backward last time (behind, maybe even foolishly so), I now want to make the case that Europe might have gotten lucky. Or, more generously: had foresight.
|
|
The reason for this reversal is Kimi K3, a Chinese model that came out last week. Open source, so free to download, usable for a fraction of the cost, and roughly on par with Fable 5.
|
|
So where's the problem, and where does this reversal come from now? Well… America has invested roughly 1,000 billion dollars in frontier models. That's a gigantic, all-or-nothing bet the entire American economy has staked its cards on. The intended business model: automate millions of white-collar jobs (lawyers, admin staff, project managers, you name it) structurally, and charge companies handsomely for it via subscriptions. The technology already works for the most part, and gets better every month, and it's genuinely impressive: if you use AI daily and build it into your work, you feel it.
|
|
But a technology being impressive doesn't mean it makes economic sense. America has to earn that 1,000 billion dollars (and then some) back to meet the sky-high expectations of investors and lenders, and to keep the country economically healthy. China? China doesn't have that problem at all. They spent only a fraction of that amount, because they copy the expensively trained knowledge out of American models through distillation (a total China move). Their AI business model isn't selling subscriptions, it's precisely about undermining the American business model. With dirt-cheap, open-source models they're wrecking prices in the market, using AI not as a product but, mercilessly, as an economic sword.
|
|
And that's the contradiction I've come to see myself: you don't need to lead if you can copy the technology and undermine your competitor. Pioneers do the hard, expensive R&D work, take all the resistance for it, and hand over the knowledge for the next generation, for free.
|
|
The good news for us? Europe, through slowness, luck, or insight (pick whichever word you believe, or like best), never joined that gigantic bet. And because of that, ironically, we might just be the ones laughing last. We get to ride along on dirt-cheap, high-quality AI technology….without ever having to go into debt up to our ears for it first.
|
| |
| |
| |
|
THE LIST · TEN THINGS HAPPENING IN AI LAND
|
|
1. Google spent more than it earned, for the first time. The perfect afterthought to the main story about the giant AI bet: Google (Alphabet) reported that, due to sky-high AI investments and capital spending, its free cash flow turned negative for the first time. The mega-corps are pouring hundreds of billions into AI infrastructure, but currently spend more on AI than it operationally brings in. Yahoo Finance
|
|
2. A bike ride made from a flat map. A clip made from a Google Maps screenshot: one flat map image in, one prompt, and out comes a first-person cycling video that exactly follows the marked route; as if you're riding through it yourself. The clever part is that the model has to understand a static map, a geographic route, and realistic motion all at once, and stitch them together. Instagram
|
|
3. VisionClaw turns smart glasses into an AI agent. Think Meta Ray-Ban smart glasses: the feed goes to Gemini Live, which plugs into an agent layer (OpenClaw) that then actually does something in the world, in real time. Not a chatbot: an AI that sees and acts. Unsettling and privacy-invasive, but fascinating at the same time. Instagram
|
|
4. NameThatUI is a picture dictionary for UI elements. Type the vague name you can't remember ("that thing with the dots") and get the right word and an example back. With AI you can now build almost anything (websites, graphic elements, interfaces), but finding the right terminology remains hard for regular people (myself included). Exactly the gap this tool closes. namethatui.com
|
|
5. Record your screen, and Claude learns the task. Anthropic launched "Record a Skill" in Claude Cowork: you record your screen while doing something, narrate what and why out loud, and Claude turns that recording into a skill it can repeat on its own afterwards. No more prompt and skill engineering, just show it. YouTube
|
|
6. Paid software gets rebuilt for free within weeks. I love Wispr Flow, talking to an AI now feels more natural than typing, except it's yet another expensive subscription. But tools like Claude Code are now used to rebuild that kind of SaaS product into free, open-source alternatives, in no time. Enter OpenWhispr. Want to install it? Copy the GitHub page's URL into Claude Code and say: I WANT THIS!!!!! GitHub
|
|
7. Jack Dorsey takes on Slack. Dorsey, co-founder of Twitter and Square, launches Buzz: group chat, but built for teams and their AI agents in the same channels. Slack, with your AI colleague as the foundation. TechCrunch · video
|
|
8. Anthropic runs its own 24/7 lofi radio. "Claude FM" on YouTube: music to think and build to, made and curated by musicians (520K subscribers). I thought it was cute, but it's also just kind of slop. YouTube
|
|
9. Season 2 of my Scorito agent: the Dutch Eredivisie. No slop here, just football. At the World Cup my agent won its friend group's pool, beating 98.6% of 1.44 million players in a forty-day sprint. Now the same agent runs the nine-month Eredivisie marathon: building a squad and predicting results. €48M budget spent, every decision logged. (Disclaimer: not financial or betting advice.) Follow the Scorito agent
|
|
10. Crossword puzzles with AI. Last Saturday I wondered: can AI make good, visually-driven crossword puzzles? Answer: yes, it can. If you're up for a challenge (it's genuinely hard) and know your Feyenoord (my great love), The Sopranos, Star Wars, or Game of Thrones: play CrossMe. Play CrossMe
|
| |
|
AGENDA
|
| |
Questions, or a tip for the next edition? Just reply to this email, I read everything.
This is the quick-scan. My longer stories on music & culture live on Substack.
You're receiving this because we had contact via marcelalexander.ai. Unsubscribe any time, with one email.
Marcel Alexander Wiebenga · geen woorden maar daden
|