When I first started college, I wanted to be a programmer. Not the “casually hacking like I was in the movie Hackers”, either. I had the whole thing storyboarded out, the one where I commute into San Francisco for some big tech company or a hot startup and make a frankly ridiculous amount of money. The 90s were absolutely stuffed with magazines and commercials glorifying that exact guy, and I was sure, so very sure, I was gonna be him.

Programming turned out to be way harder than I bargained for, at least for me. I could not get anything to work past simple text prompts, and since I was bad at math then (and, full disclosure, remain bad at math now), all the deeper concepts just sailed right over my head and waved on the way by. After two wasted semesters and many a serving of cafeteria cheese fries (no regrets, baby!), I switched to something way more my speed: multimedia.

I already loved to draw, so the digital version of creativity was right up my alley, and I went in deep. Graphics, video editing, audio production, illustration, web design. I ate all of it up. And if I can say so without my head swelling too much, I got pretty good.

But, pride comes before the crash.

“Pretty good,” meet everybody else

Naturally I figured, “hey, multimedia, I’ll get a job easily.” Then I met the job market.

Turns out the thing about being pretty good at something is that a whole lot of other people are also pretty good at it (surprise, surprise), a fair number are flat-out better (way better), and every last one of you is fighting over the same handful of jobs. As good as I thought I was, I was one face in a very big crowd, and the outcome I’d been dreaming about did not show up.

So I did what you do when you gots to eat. I went and found a job.

The college I’d graduated from needed help in the computer lab, and I needed income, so I took a gig as a lab technician. Figured it was a stopgap. Welp. Little did I know that one little decision was about to march me down a years-long road through IT that I never once planned on. College lab, then a computer repair shop, then tech support at a middle school, then a promotion into a school district’s IT department, then two more districts, plus a stint right before the pandemic (there were even a couple years of STEM/STEAM curriculum planning crammed in there somewhere, but that’s a story for another day).

The different kind of creative

By the end of all that, I was burned out. Crispy.

I did a little educational admin work to catch my breath, and here’s where I learned something I didn’t have words for at the time. You can technically be “creative” in a job like that. You find a slick new way to reformat a spreadsheet, or you talk your way around a department problem, and sure, that’s a kind of creative. But it’s not the fun kind. At least for me, anyhoo.

There’s a real difference between work that uses your creativity and work that actually feeds it, and you don’t fully feel that difference until you’ve spent a few years on the wrong side of it. I had. My brain was starving for the other kind, the kind that gives something back.

So I started feeling my way toward it. Some freelance video work. A little social media passion project I’ll tell you about another time. But it was clear my brain wanted a brand new outlet, not an old one dusted off and plugged back in.

That’s around when I stumbled onto two things that clicked the whole season into place.

The two things

First was AI.

I won’t get too in the weeds here, because I’m not looking to start a debate, and however you feel about it, I respect it. For my purposes, it turned out to be a great assistive tool for pretty much anything creative. Is it any good at creating things itself? Nope, and that’s a job I’ll never hand it. But as an assistant? For what I need, it works great.

(Quick aside, while I’m here. AI makes coding so much easier now, and part of me does wish it had existed back when I was flunking out of programming. Although, let’s be real with ourselves. I almost certainly would’ve let it do all my hard programming work for me, which in my case was, hrmm, all of my programming work. So I’d have learned absolutely nothing and called it a degree. C’est la vie.)

The second thing was the big one. Writing.

In some ways I’d always been into writing without ever clocking it as a thing I did. I even kept up a serious Yelp review streak for a couple years, which I’m now choosing to count as relevant experience. But doing it professionally? Never crossed my mind. Then I looked up and realized that between ghostwriting, newsletters, and educational email courses, writing had quietly become more fun than it’s ever been. Lucrative, too.

And here’s the part that gets me. I spent years chasing the lucrative creative job on purpose, twice, and missed it both times. The one that finally worked is the one I wasn’t chasing for the money at all. I just liked doing it.

I feel like a creative again

Now here I am, constantly noodling on how the latest update to whatever AI tool I’m using can make my workflow more efficient. Not do the writing for me. Just help me have a good time figuring out the work. After everything, that’s the part I didn’t expect to get back: the having-a-good-time part.

And don’t get me wrong, I’ve got no regrets about the detours. The programming flop (again, no regrets on the cheese fries), the design grind, even the IT adventure I try not to think about too hard; all of it went into building whoever I am now. I needed every wrong turn to recognize the right one when I finally tripped over it.

But do I ever want to go back to work that doesn’t let me flex the creative muscles? To the spreadsheet kind of clever, the kind that leaves you empty? Not a chance. I went the long way around to figure out the difference, and now that I know it, I’m not unknowing it.

I’m having way too much fun over here. Best believe though peeps, I’m planning to keep it that way.

Did you enjoy this article? If so, please consider sharing this and leaving a comment below, to share your thoughts on your creative journey. Hope to hear from you!

Share

Subscribe now

This edition is sponsored by Nvidia (because apparently, everything is!… just kidding). Computex happened in Taipei this week, which means leather jacket man reminded the entire tech space that it’s trying to run the whole game. There’s so much Nvidia news that I had to make a list to keep track.

The short version: Nvidia wants to be inside your laptop, on the side of your house, at the center of Taiwan’s economy, and on your dinner plate (just kidding… for now…). Meanwhile Anthropic filed paperwork to go public at a valuation that rounds up to a trillion dollars, and Microsoft used its own conference to quietly announce that it’s tired of renting its AI from other people. Woofa-doofa! Let’s get into it!

Here’s what’s covered today:

• Nvidia officially entered the PC market with the RTX Spark

• Nvidia and SPAN want to mount tiny AI data centers on the outside of regular homes

• Anthropic filed confidentially for an IPO at a ~$965 billion valuation

• Microsoft built its own coding AI to lean less on OpenAI and Anthropic

• Nvidia pledged $150 billion a year to Taiwan and showed off a supercomputer the size of a fridge

(I know, seems like a LOT of Nvidia news. Just to be real here, I am in no way affiliated with them. Love ‘em or hate ‘em, they DOMINATED the news this week.)

Nvidia and SPAN want to bolt a tiny data center onto your house

The story: This one sounds made up, and it is very much real. Nvidia is partnering with a smart-electrical-panel startup called SPAN, plus homebuilder PulteGroup, to mount cabinet-sized mini data centers on the outside of regular homes. They’re called XFRA nodes, and the idea is to run AI workloads using the spare electrical capacity your house isn’t using, while knocking money off your power bill. The first 100 pilot units go onto new PulteGroup homes in Q3 2026.

What happened:

• Each XFRA box reportedly holds 16 Nvidia RTX Pro 6000 Blackwell GPUs, 4 AMD EPYC CPUs, and 3TB of RAM, all liquid-cooled, in a unit about the size of an HVAC condenser.

• The hardware is built by Dell and serviced by SPAN, and it mounts outside next to your AC and electrical gear.

• The trick is that the average American home only uses about 40% of its electrical capacity, so XFRA runs compute during the windows when your house is drawing little power.

• Homeowners get a premium smart panel, battery backup, sometimes solar, and discounted (occasionally free) electricity and internet in exchange for hosting a node.

• SPAN claims it can deploy 8,000 of these units roughly six times faster and at one-fifth the cost of building a comparable 100-megawatt data center.

AI runs on sweet, sweet electricity, and the grid can’t keep up. U.S. data centers ate more than 4% of the country’s entire electricity supply in 2024, and that figure could pass 9% by 2030. Building a new power-hungry data center can take over a decade once you count the wait for grid hookups. SPAN and Nvidia found a way around that wait: treat thousands of houses like one big distributed computer, using power that’s already sitting there unused. Think of it like carpooling, except the empty seats are spare amps in your breaker box and the riders are AI workloads (and not a dummy you’re using to scam the toll booth, though some AI critics may argue otherwise).

If you’re a homeowner, this is one to keep an eye on, especially if you’re buying new construction in the Southwest where the pilots are landing. The pitch is free or cheap power in exchange for a humming box on your wall. Whether that trade is worth it is going to depend entirely on the fine print, so read it carefully before you let a chipmaker move into your side yard. And let’s not forget: these may become tempting targets for copper thieves AND those with the anti-AI sentiment.

Nvidia is now coming for your laptop too

The story: On May 31 and June 1 at Computex, Jensen Huang officially unveiled the RTX Spark, an Arm-based superchip for Windows laptops and small desktops. This is the chip the rumor mill spent a year calling “N1X” and “N1.” It packs a 20-core Nvidia Grace CPU and a Blackwell GPU onto one piece of silicon with up to 128GB of unified memory, and it ships this fall in PCs from Dell, HP, ASUS, Lenovo, MSI, and Microsoft.

What happened:

• Huang said this on stage, and he wasn’t being coy: “A long time ago Nvidia used to be a GPU company.” YOU DON’T SAY.

• More than 30 laptops and 10 desktops are slated to launch this fall with the RTX Spark inside.

• Huang called the move the biggest reinvention of the PC since the smartphone reinvented the phone.

• Nvidia stock jumped over 6% on the announcement, Dell climbed over 10%, and HP gained 8%.

• The chip is built to run “agentic AI” locally, meaning AI agents that keep working in the background without phoning home to the cloud every two seconds.

For about two decades, the Windows laptop world has belonged to Intel and AMD, with Qualcomm elbowing in on the Arm side. Now Nvidia walked in carrying the biggest AI-chip business on the planet and proclaimed that it was all that and allllll the bag of chips (figuratively and literally, if you think about it). The reason this matters is the design philosophy underneath. These chips are tuned to run AI locally on your machine, which is the same direction Apple has been pushing with its M-series silicon. A real three-way fight over the future of the personal computer just broke out, and the prize is which company gets to define what an “AI PC” even means (meanwhile, Skynet could not be reached for comment).

If you were planning to buy a Windows PC in the next year, it might be worth holding out until fall to see the first RTX Spark machines, or at least to see what the new competition does to the price of everything else. Will these new PCs be winners or wieners? Stay tuned peeps. It’s gonna be a show, fa-sho!

Anthropic filed to go public at nearly a trillion dollars

The story: On June 1, Anthropic (the company that makes Claude) told the government it wants to start selling pieces of itself to the public (aka errbody can get a slice ‘o the pie, as long as they got the cash to do so). Right before this, Anthropic raised a fresh $65 billion, and that money raise put a price tag on the whole company of about $965 billion. They’re basically $35 billion away from being worth a trillion dollars, and you can’t even buy the stock yet.

What happened:

• That $965 billion price tag is now bigger than OpenAI’s, which sits around $852 billion. Anthropic quietly passed its biggest rival.

• The company is reportedly on pace to bring in about $47 billion a year, up from around $10 billion a year ago

• The $65 billion came from a who’s-who of big investment firms (Sequoia, Altimeter, Dragoneer, and a bunch more).

• On top of that, two firms (Apollo and Blackstone) are lending Anthropic another $36 billion, which pushes the total pile of new money close to $100 billion.

• Anthropic isn’t alone. Elon Musk’s SpaceX is also lining up to go public this week, and OpenAI is expected to follow soon. Everybody’s heading for the stock market at once.

A few weeks back I mentioned that Sam Altman and Elon Musk were both pointing their companies toward the stock market. Three AI heavyweights (Anthropic, Elon Musk’s SpaceX, and OpenAI, the company behind ChatGPT) are all about to let everyday people buy in, around the same time. That’s a big deal, because the stock market is the most public scoreboard there is. If people throw cash at the thing and the prices climb, it’s a giant vote of confidence (in Anthropic and AI, in general). If the prices flop, welp, might not be the greatest look for these intelligences that are artificial.

Of note is how fast this all happened. Anthropic raised $65 billion and filed its “we want to go public” paperwork in the same week. A year ago, a company raising a single billion would’ve been front-page news. The fact that this barely raises an eyebrow now is the part that should actually raise your eyebrow.

Microsoft built its own coding AI because why not

The story: On June 2 at Microsoft Build 2026, Satya Nadella unveiled Project Polaris, Microsoft’s homegrown AI coding model. Polaris is set to replace GPT-4 as the reasoning engine inside GitHub Copilot by August. At the same conference, Microsoft also made Anthropic’s Claude a first-party option inside Azure AI Foundry, sitting right next to OpenAI, DeepSeek, Llama, and Mistral.

What happened:

• Microsoft has spent the AI era as the cloud provider and the deep-pocketed investor in other people’s models, first OpenAI, then Anthropic.

• Polaris is its first serious crack at a proprietary in-house model built to go toe-to-toe with the frontier labs.

• Nadella framed the shift as AI moving from “synchronous assistants” to “async coworkers that can execute long-running tasks.”

• Agent Mode is now the default across Microsoft 365 Copilot, and the Windows Agent Framework was open-sourced under the MIT license.

• The Pentagon also handed Microsoft the lion’s share of a $9.69 billion software consolidation contract. Good week to be Redmond.

Here’s what’s actually going on underneath the announcement, in plain terms. Microsoft is one of the biggest buyers in the world of AI tokens, which are the units of work an AI model does every time you ask it something (like tokens that peeps used to plunk into arcade machines, making many a kiddo’s allowance disappear in an instant). Every prompt run through Copilot costs Microsoft money it pays out to model providers. When your bill runs into the tens of billions a year, eventually you start wondering why you don’t just build your own AI.

When a $3 trillion company looks at its AI costs and decides, “why for we not do that yesterday? DO NOW,” it’s a reason to take notice. Most people and small businesses won’t train their own models, obviously. The principle still holds: the more an AI tool becomes core to how you work, the more it pays to understand what you’re actually spending and whether there’s a leaner way to get the same result. Free trials are great for kicking the tires. The long-term bill decides if you’re gonna drive the thing.

Nvidia (hi, again) pledged $150 billion a year to Taiwan and showed off a fridge-sized supercomputer

The story: Back at Computex, Huang said the quiet part out loud: Nvidia plans to spend as much as $150 billion a year in Taiwan, and it’s opening a new Taipei headquarters. Taiwan, in his words, is the “epicentre of the AI revolution.” The centerpiece of Nvidia’s booth was the Vera (which we talked about last week) Rubin NVL72, a rack-scale supercomputer that just swept multiple Best Choice Awards at the show for the new “AI factory” category.

What happened:

• “Many years ago we had 10 partners. Five years ago, maybe 50 partners. Now we have 150 partners,” Huang said about Nvidia’s Taiwanese supply chain.

• AMD’s Lisa Su also showed up and pledged over $10 billion of her own in Taiwan investment, so this is bigger than just Nvidia.

• Vera Rubin NVL72 is the rack-scale machine behind the “AI factory” concept, which is Nvidia’s pitch for purpose-built data centers that crank out AI models at industrial scale.

• The new Taipei HQ adds a long-term anchor to a relationship already moving roughly $150 billion a year through the same supply chain.

• The show had basically zero cryptocurrency content this year, which tells you exactly where the Taiwanese tech ecosystem has thrown its chips.

Last week, I wrote that Nvidia was DoorDashing its newest chips to four big companies. This week the same company committed to spending more annually in one island than the entire GDP of most countries. The relationship has crossed into something that looks a lot like industrial policy run by a private company. When one firm decides to drop $150 billion a year in one place, the geopolitics around that place shift, the supply chain bends to fit, and every competitor has to draw up plans around the new center of gravity. Might have to update world maps to “Taiwan (aka Leatherjacketmanland), Sponsored by Nvidia.”

Whoever wins the AI model race, the company selling the hardware underneath all of them gets paid either way. That’s the position Nvidia is in, and this week it spent like a company that is desperate to stay there. The hardware roadmap they laid out runs well into 2027. For anyone building on top of AI, the thing to remember is that the compute you depend on is set to keep getting faster and cheaper for a while, which is worth factoring into whatever you’re planning. Worth watching, too, is who (if anyone) ever manages to challenge that position, because a market with one dominant supplier tends to get expensive for everyone downstream eventually.

What ties it all together

This was a week of enormous institutional moves. Nvidia wants in on the PC pie, floated the idea of turning houses into compute nodes, and pledged a country-sized budget to Taiwan. Anthropic took the first real step toward becoming a public company. And Microsoft made its declaration of partial independence.

Step back and look at the shape of it. The AI industry has gotten big enough that it’s now reorganizing every industry around it. Personal computing is being redesigned around AI agents. The power grid is being re-plumbed to feed AI compute. The public markets are about to get their first honest read on AI valuations. The world’s biggest software company is building its own models so it can stop paying the AI fun tax. The common thread is that AI stopped being a product you use and became infrastructure that everything else now has to bend around. Scary or innovative? Perhaps both? I’ll let you, the reader at home, decide.

What I’m watching

• Whether the RTX Spark holds up on performance once reviewers get the first laptops this fall

• Whether anyone actually signs up to host an XFRA node, and what the fine print really asks of them

• How the SEC review of Anthropic’s S-1 unfolds, and whether OpenAI files its own paperwork before the end of June

• Whether Microsoft’s Polaris truly replaces GPT-4 inside Copilot by August, or quietly slips

• Whether the next big memory headline is a price drop or another shortage warning

  

Thanks so much for reading this edition of The Weekly Think.

See you next week, leather jacket man megafans fellow thinkers!

Did you enjoy this week’s Think? If so, please consider sharing this and leaving a comment below, to share your thoughts on these latest AI-related rumblings. Hope to hear from you!

Share

Subscribe now

The year was 1996, and my family could not afford a computer. I wouldn’t get one of my own until college. But my local library, in a move I am forever grateful for, wired all of its machines to the ‘ol (new??) world wide web, and just like that, the door was open.

And man, did I get lost in the sauce.

I discovered GeoCities. I chatted ALLLL the time at a place called the Tokyo Tea Room. I looked up Game Genie codes for games I was definitely going to beat that weekend, and I devoured every scrap of gaming and anime news I could find. It was an addiction in the worst way, my guys, and I loved every single second of it.

Somewhere in all that loafing, surrounded by webpages other people had built, I landed on the thing I actually wanted to do. I didn’t just want to read these pages. I wanted to make my own.

The embarrassing dilemma

Here was the problem. I had a free GeoCities account, which meant I had my own little plot of internet land just waiting to be built on. What I did not have was a computer at home, and home was where I was at my most creative.

Now, the obvious solution here is for someone to lean over and say, “just design them at the library, dummy.” And that’s fair. That’s completely fair. But I was a teenager with unsupervised internet access for the first time in my life, and I was far too busy having the time of my life to do anything as responsible as using my computer hours productively.

So the question stood, and it was for real. How was I supposed to design webpages with no computer to design them on?

My solution was to draw them. In a notebook. By hand.

What was actually in those notebooks

I loved my notebooks back then. Beyond the occasional journal entry, they were stuffed with my random thoughts and a whole lot of sketches, because I was a teen who loved to draw (I still love to draw, but back then I really loved to draw).

As I fell into the internet age, my notebooks fell right in with me. I was copying down cheat codes and gaming news by hand, because the library didn’t have printers yet either. But I was also doing something a little wild. I was sketching out my dream webpages in full, complete with underlined links, hand-drawn logos, and web banners (and, y’all, web banners were a HUGE deal in the GeoCities era, a load-bearing pillar of any self-respecting homepage).

I was designing like a maniac. It was a completely computerless kind of design, with no way to ever actually publish any of it, and I poured myself into it anyway.

That’s the part I want you to sit with for a second, because I didn’t see it clearly myself until recently. I loved the designing so much that I did it with no computer, for no audience, with no way to ship the thing. The making was the entire point. There was no other point. Just me, a pencil, and a notebook full of wild webpage designs.

The candy store, then and now

The thing about the early internet that’s hard to explain to anyone who missed it is the sheer daily wonder of it. Something new and amazing showed up every single day, people were innovating in every direction at once, and the rest of us got to just stand there and watch it happen in real time. You were the kid in the candy store, except the manager had leaned over and said, “bet you can’t eat as much candy as you want, for free.” Welp. Challenge accepted, sir. All day, every day.

I bring this up because, fast forward to 2026, AI right now feels an awful lot like that. The chatbots update on what feels like a constant cycle, the services are generating every kind of content you can imagine, and the thinking models are getting pointed at real problems and actually making progress on some of them. It’s that same sense of something new showing up before you’ve finished marveling at the last thing.

But there’s one big ‘ol difference in this modern era. Back in the 90s, nobody was standing over the new creations calling them garbage. Today, a whole subset of people have decided that anything made with AI gets labeled “slop,” and the heat behind that critique runs hotter than anything I remember from the early internet days. I’m not here to settle who’s right about that. I’ll just note that the wonder and the contempt are sharing the same room right now, and leave it there.

Why I won’t hand the machine the whole thing

So where do I stand on this whole thing?

I don’t mind using AI as an assistive tool one bit. A good amount of what I make these days has AI somewhere in the process. But using it as the one and only tool, the lone artist doing the actual creating? I don’t do that, and not because some online tribe told me not to.

I come from a traditional multimedia background (Macromedia, anybody?… uhh… hello?), and even before that, I was the kid drawing webpages in a paper notebook for the pure joy of it. I love the design process far too much to just hand it over to a machine. The thing AI is offering to skip for me is the exact thing I loved then and still love now.

And that’s the part I keep coming back to. The notebook kid would have been genuinely baffled by the offer. Picture it. You walk up to a teenager who is so in love with designing that he’s doing it on paper, with no computer, for an audience of nobody, and you tell him you’ve got a machine that will do the whole thing for him instantly. He doesn’t want that. He would look at you like you’d offered to eat his Hostess snack cake on his behalf (and ain’t NOBODY eating MY Hostess snack cake). The drawing wasn’t the obstacle between him and the webpage. The drawing was the entire reason he was sitting there.

I’m still that kid. The offer hasn’t gotten more tempting just because the machine got better at accepting it.

I’m bringing the notebook back

As I sit here typing this, I keep thinking about that kid sketching out fanfiction-level webpages he had no way to build, designing with the kind of focus that would’ve made you think he was Doug Funnie with a library card (and yes, Doug eventually got internet episodes, but we don’t talk about those middle-school-era ones that hardly anybody I know watched).

My design process has changed many a time since then. But I don’t think the kid was wrong about anything, and the fear I keep hearing, that AI is going to cost people their creativity and their originality and their authentic self, would not have scared him one bit. It can’t take any of that from you unless you hand it over. I know, because that whole creative fire was burning years before I ever touched a tool, fueled by nothing but a library card, a spiral notebook (and a frankly irresponsible number of Hostess snack cakes).

I kinda wanna break out the notebooks again. Not as some productivity trick, and not as a protest against anything. It’s because that kid at the library understood something I’ve spent thirty years occasionally forgetting, which is that the wonder was never coming from the tool. Not the computer I didn’t have, and not the AI I do. It was coming from me, hunched over a hand-drawn graphic I’d never publish, having the time of my life.

The pencil didn’t make me creative. Neither will the machine. That part was always mine to keep.

Did you enjoy this article? If so, please consider sharing this and leaving a comment below, to share your thoughts on your creative process. Hope to hear from you!

Share

Subscribe now

Greetings, fellow humans! It’s your totally human boy, Eric. Last week was all about the money, talent, and lawsuits (in a way, it always sorta is 😵‍💫). This week the conversation got a little bigger, and by bigger I mean a religious leader showed up with thoughts. Pope Leo XIV published the first papal letter in history dedicated to artificial intelligence (though imagine if a pope in the Byzantine times did that), and words were not minced.

The rest of the week kept the theme going. Anthropic planted a flag on the ad question we talked about last time, the memory crunch finally came for a popular PC gaming handheld, an AI model solved a math problem humans had been stuck on, and Nvidia started hand-delivering its newest brains (not Jensen Huang’s brain, to be clear) to the biggest labs. Grab a hot dog, dawg. Let us to begin, humans.

Here’s what’s covered today:

• Pope Leo XIV published the first-ever papal encyclical on AI, with Anthropic on stage

• Anthropic promised Claude will stay ad-free (one week after ChatGPT turned ads on)

• The Steam Deck jumped up to $300 thanks to “RAMageddon” (thanks, Jensen)

• An OpenAI model disproved a longstanding math conjecture

• Nvidia hand-delivered its new Vera chips to Anthropic, OpenAI, and SpaceX (thanks, Jensen?)

Pope Leo XIV wrote 42,000 words on AI, and Anthropic was in the room

The story: On May 25, Pope Leo XIV released Magnifica Humanitas (Latin for “Magnificent Humanity,” not to be confused with my favorite emo rap album of 2005), the first encyclical of his papacy and the first one in Roman Catholic Church history built entirely around artificial intelligence. An encyclical is intended to be an authoritative kind of letter a pope writes, the heavy-hitter format usually reserved for huge moral questions. He picked AI for his first one, which is a notable choice given everything else a new leader could lead with.

What happened:

• The letter runs 42,300 words and took ten years of Vatican dialogue with the tech industry to produce.

• Leo signed it on May 15, exactly 135 years after Pope Leo XIII’s famous 1891 letter on workers’ rights (the name match is on purpose).

• He argues AI should lift up human dignity and the common good, and warns against a pure race for “more powerful algorithms and larger datasets” driven by profit or geopolitical muscle.

• The Vatican launch featured Anthropic co-founder Chris Olah, who leads the team that studies what’s actually happening inside these models.

• One of his notable thoughts: he calls for AI to “help, not hinder” the development of kids.

What makes this newsworthy has nothing to do with whether you give Pope Leo XIV’s words any weight. It’s the signal. A 2,000-year-old institution with over a billion adherents spent a decade in conversation with the tech industry, then put an AI researcher on stage to mark the occasion (the religious leader & an AI man, sharing the stage, sounds like something from an episode of Futurama). When an organization that size and that old decides AI is the subject of its first major statement, a lot of people are going to take their cue from it.

That’s the real story here: AI has officially become big enough that the institutions shaping how billions of people see the world are now staking out positions on it. Governments, courts, and now the largest religious body on the planet. The conversation has left the tech bubble, and that means the rules, norms, and pressures around this technology are about to start coming from places that have nothing to do with Silicon Valley (personally waiting on AI takes from Mayor McCheese of McDonaldland).

Anthropic said Claude stays ad-free (one week after ChatGPT turned ads on)

The story: Remember last week, when I told you OpenAI flipped on a self-serve ad platform inside ChatGPT and basically said “you’ll have both and you’ll like it” to the charge-or-advertise question? Well, this week Anthropic walked up to the same fork in the road and went the other direction. The company published a piece explaining that Claude will stay ad-free, on purpose, for the foreseeable future.

What happened:

• Anthropic’s argument is that advertising incentives pull against being a genuinely helpful assistant.

• The worry is simple: once an ad pays the bills, there’s pressure to keep you scrolling and nudge what the assistant says.

• Anthropic says it plans to widen access to Claude without leaning on ads to do it.

• This lands one week after OpenAI announced it’s chasing $100 billion a year in ad revenue by 2030.

So now we’ve got a real split in the road, and it’s a fun one to watch. One of the two biggest AI companies is copying the Google playbook (don’t tell coach, or it’s 100 laps). The other is betting people will pay for a tool that has zero incentive to keep them glued to the screen. It’s the difference between a free newspaper stuffed with inserts (though Sunday newspapers used to be fire) and a subscription you actually chose. Two takes (that is to say, OpenAI’s and Anthropic’s) on what you want from the thing you ask questions all day.

The practical takeaway is to notice which one you’re using and why. If a tool is free, ask who’s paying for it. That habit is just good media literacy, and it matters more every week as these assistants get woven into how we all work.

The Steam Deck got up to $300 pricier, and AI is the reason

The story: Last week I walked you through Samsung crossing $1 trillion on the back of the memory boom, and I mentioned the bill was eventually coming for the rest of us. Welp, gamers, here we are. On May 27, Valve raised the price of its Steam Deck OLED handheld by as much as $300, blaming rising memory and storage costs directly. The internet has started calling this whole mess “RAMageddon,” which yeah, accurate.

What happened:

• The 512GB model went from $549 to $789, and the 1TB model jumped from $649 to $949.

• Valve was clear the hardware didn’t change at all; only the component costs did.

• The 1TB Steam Deck now costs more than a PS5 Pro, which is a sentence that would have sounded insane a year ago.

• Valve is far from alone: Sony, Microsoft, and Nintendo have all raised console prices recently for the same reason.

• The root cause is the AI data center boom eating up the world’s memory supply (one report says OpenAI alone reserved roughly half of one chipmaker’s entire 2026 RAM output).

Here’s the through-line that connects this to everything else in the newsletter. The same AI buildout minting trillion-dollar chip companies is the reason your gaming handheld costs as much as a laptop now (if a Game Boy had cost as much as a PC in 1989, there would’ve been RIOTS). The memory that would have gone into affordable consumer gadgets is getting routed to data centers instead, because we all need data centers because leather jacket man (aka Jensen) said so. It’s one giant game of musical chairs, and right now the AI labs have bought most of the chairs.

If you’ve been eyeing a Steam Deck, a gaming PC, or honestly any device with a lot of RAM in it, the practical move is to buy soon or buy refurbished. Prices aren’t expected to ease until new memory factories come online, and those are a 2027-or-later situation (might as well call it Cyberpunk 2027). This is the rare case where waiting probably costs you extra.

An AI model solved a longstanding math problem

The story: On May 22, OpenAI announced that one of its models disproved a longstanding conjecture in discrete geometry, a corner of math that deals with shapes, points, and how they fit together. In plain terms: humans had a hunch about how something worked, nobody could prove it either way, and the AI found the counterexample that settled it (like the cyber-nerd it is).

What happened:

• A “conjecture” is an educated guess that hasn’t been proven; disproving one means finding a clear case where it breaks.

• The model didn’t just guess; it produced a result mathematicians could check and confirm.

• This follows a similar feat where another frontier model contributed to a real math proof earlier in the spring.

• These are narrow, specific wins, not “the AI is now smarter than mathematicians” headlines.

This one is worth understanding correctly, because it shows what these tools are actually good for. The common knock on AI models has been that they only ever average out the internet and hand it back to you (sorta like how Taco Bell reconfigures existing ingredients into the latest viral menu item). A math result like this works differently. A person decided which problem was worth attacking, set up the approach, and checked whether the output held up. The machine ran a staggering number of dead-end attempts very fast, the way a calculator crunches numbers a person could technically do by hand. The insight, the judgment, and the verification stayed human start to finish.

If you work in any field with hard, well-defined problems (engineering, research, finance, logistics), here’s the part worth your attention. These tools are good at chewing through a huge pile of possibilities quickly, and they’re useless without a sharp person deciding what’s worth chewing through and whether the answer is any good. So the person who gets ahead is the one who knows which problems are worth pointing the tool at, and who can tell a real answer from a confident-sounding wrong one. The hard part was always the human judgment, and it still is.

Nvidia hand-delivered its newest chips to the biggest labs

The story: Leather jacket man spent the week reminding everyone who actually supplies the shovels for this whole gold rush. Jensen Huang says the company’s new Vera CPU is unlocking what he pegs as a $200 billion market for agentic AI, with $20 billion already sold this year. In fact, the first units were reportedly hand-delivered to Anthropic, OpenAI, SpaceX, and Oracle. It’s like DoorDash, but it’s Jensen and you tip him BILLIONS.

What happened:

• The Vera CPU is the “brain” half of Nvidia’s next-gen Vera Rubin platform, built specifically for AI agents that run continuously.

• Hand-delivering the first chips to your biggest customers is equal parts logistics and showing off, and Nvidia knows it.

• Nvidia’s most recent quarter came in around $81.6 billion in revenue, up roughly 85% from a year ago.

• Anthropic, OpenAI, and SpaceX (the Colossus supercomputer crew from a few weeks back) are all on the delivery list, so the same handful of players keep showing up.

The pattern here is one I keep returning to. No matter which AI lab wins the model race, Nvidia sells the chips to all of them. It’s the picks-and-shovels business, and right now it’s the best seat in the house. When you hand-deliver your newest product to four of the most important companies on earth and they’re all thrilled to get it, you’re not really sweating the competition. You’re the supplier, and the streets love you or hate you.

For the builders and the curious: watch where these Vera chips point, because they’re purpose-built for agents that can run 24/7, 365 days a year, including leap years. That tells you where the whole industry thinks the next year is headed. The bet is on AI that constantly runs, and the hardware is already being trucked to the front door.

What ties it all together

Last week the story was the business of AI: money, lawsuits, talent. This week the story zoomed out to the meaning of AI: Pope Leo XIV weighing in on human dignity, two companies splitting on whether to sell your attention, and an AI quietly solving a problem humans couldn’t.

Step back and check it out. A 2,000-year-old institution and a frontier lab shared a stage. One AI company bet on ads, the other bet against them. A gaming handheld became a casualty of the data center boom. A model cracked a real math mystery. And Nvidia kept calmly selling chips to everyone in the room. The technology is moving fast, but this week was a reminder that the bigger questions (who pays, who benefits, what stays human) are finally getting asked by the kind of people who don’t usually show up to tech keynotes.

A year ago, “religious leader writes a letter about AI” would have sounded like a headline from a satire site. This week, it was literally Monday.

What I’m watching

• Whether other AI companies follow Anthropic’s ad-free stance or OpenAI’s ad-funded one

• Whether memory prices (and gadget prices) climb even higher before those new factories open

• Where researchers crack the next hard problem using AI, and whether it's science, medicine, or math again

• When the Vera chips start showing up inside products the rest of us can actually use

Thanks so much for reading this edition of The Weekly Think.

See you next week, fellow actual humans!

Did you enjoy this week’s Think? If so, please consider sharing this and leaving a comment below, to share your thoughts on these latest AI-related rumblings. Hope to hear from you!

Share

Subscribe now

There’s a part of every creative project that few are actually excited to do. It’s the proofread you’ve already done twice, or the formatting pass where you’re moving commas around for the fourteenth time, or that third read-through where your eyes have officially stopped reading and started skimming over sentences you wrote a week ago. You know exactly the part I’m talking about, because everyone who has ever made anything has felt their brain check out somewhere around that stage of the process (though, no hate if you love proofreading, because you deserve some kind of trophy if that’s the case).

That’s the boring half. And here’s the thing nobody is really telling you about AI, because everyone is too busy arguing about whether it’s going to take over the world by next Tuesday.

AI is genuinely, almost suspiciously good at the boring half.

What it’s actually good at

I’ve worked as a creative for over twenty years, which means I’ve done my fair share of proofreading, light formatting, and tedious cleanup work. So when I say AI is good at this stuff, I’m not guessing about it. It catches the kind of small errors that vanish from your sight after you’ve stared at the same paragraph for too long (ex: “Looks like you meant to say ‘thought’, but you said ‘though’. Don’t worry champ, I got this!”). It handles light formatting work without sighing about it. It will rephrase a clunky sentence for you so you can finally see what you actually meant to say, which is honestly one of the more useful things a writing tool has ever done (for this guy who doesn’t English so well, at times).

The same goes for quick proof passes, small graphic edits, and the kind of cleanup work you do at the end of a project when your creative energy is gone and you just want the thing to be finished. AI is good at all of it, most definitely, and I see no reason to be coy about that.

Some creatives feel weird about admitting they use AI for anything at all, like admitting it costs them their membership to some sort of Creative’s Costco. Worry not, my friends: it doesn't. A calculator is useful, as is spell-check. Saying "this tool helps me power through the unfun parts" doesn't take anything away from the rest of your work. It's just a description of how a tool works.

Now, I’ve spent a good amount of time praising what AI does well. Is it perfect though? Welp…

Where it falls apart

Now here’s the other half of the story, and this is where the discourse gets the whole thing backwards.

AI is bad (not bad meaning good, like they used to say in the 90s), sometimes spectacularly bad, at the part of creative work that is actually creative. The part where you’re figuring out what the thing should even be. The part where your specific point of view, your taste, and your weird particular brain are the entire reason the work has any reason to exist in the first place.

Hand that part over to AI, and what you get back is the literary equivalent of stock footage. It’s technically functional but entirely forgettable, like a playground slide that’s only a foot high. There’s no point of view in it because there was no point of view going into it. You can feel the absence the moment you start reading, even if you can’t articulate exactly what’s missing.

This is the stuff people love to hate, and with good reason. It’s content the machine generated, that the person didn’t bother to read carefully, and that the audience scrolled past without thinking. The whole transaction happened, and the world is now slightly more cluttered than it was before, with no actual meaning added to it.

That isn’t AI’s fault, though. AI didn’t volunteer for the job. Somebody pointed it at a creative decision and asked it to make that decision for them, and AI did the best it could, and its best was a sentence that started with the words “And honestly”.

The skill is knowing which half is which

So what’s a creative rockstar like yourself to do? The real skill of using AI well is being able to look at your own work and know which half is the boring half and which half is the part that’s actually yours.

The boring half is execution. You already know what you’re trying to make, and now you have a seemingly unsurmountable grind ahead of you. Fear not though, Creative King (or Queen… or… Chief); that’s where AI earns its place, and that’s where you should be using it without guilt.

The creative half is something completely different. It’s where the decisions live. What is this thing actually about? Who is it for? What’s the specific observation, the unusual angle, the small true detail that only you would have caught? That part isn’t safe to outsource, because the second you outsource it, the thing you’ve made stops being yours, and people can tell. They might not be able to name what’s missing, but they can feel it.

The trouble most people are running into right now is that they’ve quietly flipped the equation. They’re handing AI the creative decisions and then doing the proofreading themselves, which is the precise opposite of how this should be going. It’s like paying some random dude to paint for you, then taking that painting back to your studio and trying to fix it yourself. You’re gonna have a bad time.

Take back the part you actually love

If you’ve been using AI on the wrong half, here’s the good news. The half you’ve been handing off is still right there, waiting for you to pick it back up (like that cute puppy you’ve been neglecting).

The next time you sit down to make something, try this. Let AI handle whatever part of the project bores you. Hand over the proofread, the formatting, the cleanup, the small graphical work, all of it. Don’t feel weird about giving it the grunt work, and don’t feel like you have to defend it to anyone. That part of the work was always a chore, and the tool exists to handle chores (I wish I had a tool to handle chores when I was 10, but alas…).

Then take back the part you actually love. The opening line that nobody else could have written. The angle that only makes sense because you’ve lived your specific life. The choice of which detail to include and which one to leave out. The voice that makes the thing recognizably yours when somebody reads three sentences of it without your name attached.

That’s the part you got into this for. It’s the part you’ve been quietly losing reps on, and it’s the part the machine cannot do for you, no matter how nicely you ask (*glares at the chatbot*). You don’t have to fight AI to get it back. You just have to stop handing it the part that was yours to begin with, and start using the tool for the part you were never going to enjoy doing anyway.

The boring half belongs to the machine now. The good half is still yours. Go take it back.

Share

Subscribe now

Ahoy, AI news, starboard bound! So, Google I/O came, I/O saw, and I/O dropped a whole lot of Gemini on us. I/O am intrigued (okay, that’s enough of the I/O)! The short version: Google spent the week reminding everyone that the company with two billion users and its own chips (sadly, not Ruffles) is, in fact, still here.

The bigger theme this week ran underneath the product news: people, lawsuits, and money. Anthropic landed one of the most respected researchers alive. Musk’s big swing at OpenAI got tossed by a jury before lunch. And OpenAI quietly turned on the thing everybody claims to hate but secretly keeps the lights on: ads. Let’s get it, y’all.

Here’s what’s covered today:

• Google I/O dropped Gemini 3.5, a 24/7 agent named Spark, and smart glasses

• Andrej Karpathy left the OpenAI orbit and joined Anthropic

• A jury threw out Elon Musk’s lawsuit against OpenAI in under two hours

• ChatGPT opened a self-serve ad platform (your chatbot has a billboard now)

• Samsung crossed $1 trillion, and the memory crunch is minting winners

• And your “oof” moment of the week

Google I/O: Gemini 3.5, a 24/7 agent, and a $100 price cut nobody saw coming

The story: On May 19, Google held its I/O keynote and did what Google does best lately, which is cram Gemini everywhere it can (still waiting on Gemini: The Wholesale Store and $1.25 Polish sausages). The headliners were Gemini 3.5 Flash (available that day), a 24/7 personal agent called Gemini Spark, a new design language called Neural Expressive, and a surprise price drop on its top AI plan.

What happened:

• Gemini 3.5 Flash beats the older 3.1 Pro on coding and agentic benchmarks while running 4x faster, and it’s already powering Search.

• Gemini Spark is a personal agent that runs on Google’s cloud (so it keeps working even when your laptop is closed) and takes actions across Gmail, Docs, and Calendar.

• Google AI Ultra dropped from $250 a month to $100, and the old $250 tier became $200 with the same features. A price cut, in this economy!

• The first Android XR “audio glasses” are coming this fall, built with Samsung and Qualcomm, with frames by Warby Parker and Gentle Monster.

• The glasses pair with both Android phones AND the iPhone, which is a quiet little flex aimed directly at Apple.

Last week, I wondered whether I/O would land a full Gemini 4 or settle for an incremental bump. The answer was sorta neither. Google shipped a 3.5 (“like 3, but with a .5!”), but the real story was Spark. A model is a tool you pick up and put down. An agent is something that keeps working while you sleep. Your phone running little, tireless virtual contractors; Google just made that the centerpiece.

If you use Google’s stuff all day (fair, in this day and age), keep an eye on Spark when it lands for Ultra subscribers next week. The price cut to $100 tells you Google wants this in as many hands as possible, as fast as possible.

(Hrmm. I feel like I missed an Old MacDonald joke opportunity here. Oh well… E, I, E-I/O! Felt good to get that out of my system.)

Andrej Karpathy left the OpenAI world and joined Anthropic

The story: On May 19, Andrej Karpathy announced he’s joining Anthropic. If that name doesn’t ring a bell, here’s the context: he was a founding member of OpenAI, he ran Tesla’s self-driving AI program, and he’s one of the most respected teachers and researchers in the entire field. People learn how neural networks work from his free YouTube lectures. In the AI world, my mans is considered all that and a bag of chips.

What happened:

• Karpathy is starting a team focused on using Claude to speed up its own pre-training research (AI helping build AI; the snake is eating its tail, and it’s thrilled about it).

• Pre-training is the giant, wildly expensive phase that gives a model its core smarts.

• Anthropic also picked up Chris Rohlf, a 20-year cybersecurity veteran from Meta, for its frontier red team.

• This caps a stretch where Anthropic has become a magnet for top talent and acquisitions (it bought developer-tools company Stainless the day before, on May 18).

Here’s why this is more than AI tea. Karpathy could have gone anywhere (or did what Michael Jackson commanded: to be startin’ somethin’ for himself), and he picked the company betting that AI-assisted research, not just raw compute, is how you stay at the frontier. Think of it like a head chef leaving the biggest restaurant in town to go work at the small, indie kitchen down the street. People notice. Other chefs start updating their resumes. Talent follows talent, and right now the talent is pointing at Anthropic.

If you’re early in your own AI learning journey, check out one of Karpathy’s “Zero to Hero” lectures. The guy explains hard things simply, which is exactly how to play this whole AI game.

A jury threw out Musk’s OpenAI lawsuit in under two hours

The story: WELP, the Musk legal saga got its own ending this week (for now, at least). On May 18, a federal jury in Oakland (shout out to the Bay Areaaaaa) took less than two hours to throw out Elon Musk’s lawsuit against OpenAI and Sam Altman. They didn’t even break a sweat.

What happened:

• Musk had accused Altman and OpenAI of “stealing a charity” by turning the nonprofit into a for-profit operation.

• The jury never reached that question. They ruled Musk simply waited too long to sue (the legal term is “statute of limitations,” which is a fancy way of saying “you snoozed, you lose”).

• The verdict was unanimous, and the judge agreed with it.

• Musk says he’ll appeal, calling it a “technicality.”

• Fun footnote: this same week, the new coding tool Cursor shipped a model partly trained on Musk’s Colossus 2 supercomputer, which is the sequel to the Colossus 1 that Anthropic leased last week. Elon’s compute is everywhere, even when his lawsuits aren’t winning.

Both Altman and Musk are steering their companies toward the money right now, in what could be record-breaking IPOs. Not a great time to be handed messy verdicts. For Altman, the timing is a gift. For Musk, it’s a day he’d probably like to forget.

The wild reality of it all: the biggest fights in AI are increasingly happening in courtrooms and boardrooms, not research labs. It’s Three O'Clock High up in here and everyone’s coming to see Jerry VS Buddy.

ChatGPT now sells ads (aka here’s your report, now how about some juicy Chicken Nuggets?)

The story: On May 21, OpenAI launched a self-serve advertising platform inside ChatGPT, basically an Ads Manager where companies can buy, run, and tune ad campaigns (“because if Google can do everything, so can we!”, said Altman in a press release I just now imagined). OpenAI is reportedly targeting $2.5 billion in ad revenue this year and a frankly bonkers $100 billion a year by 2030.

What happened:

• The platform supports both pay-per-view and pay-per-click ads, the same models that built Google and Facebook.

• It plugs into the big ad agencies and ad-tech firms (Dentsu, Omnicom, Publicis, WPP, Adobe, and others).

• OpenAI swears ads won’t influence ChatGPT’s actual answers, only where ads appear around them.

• This pairs with a separate launch on May 15, where ChatGPT added a personal finance dashboard that connects to your bank accounts.

Every free product you’ve ever loved eventually meets the same fork in the road: charge everybody, or sell ads. ChatGPT just said “why not both?” And yeah, it makes sense. Running these models costs a ludicrous amount of money, so somebody has to pay the piper. The interesting question is whether people will tolerate ads inside a tool they treat like a trusted advisor (remember the Claude Super Bowl ads?).

Here’s the practical bit for you: if you market anything, a brand-new ad channel just opened up, and the early movers on a new channel always get the cheap attention before everyone else floods in. Worth a look-a-roo if you rely on search engine ads.

Samsung hit $1 trillion, and the memory crunch is minting winners

The story: Last week I talked about the memory shortage and the doom of rising laptop prices. This week, here’s the flip side of that same, painful coin. On May 21, news broke that the AI memory boom pushed Samsung’s market value past $1 trillion. The same shortage squeezing your wallet is making the chip companies Dr. Evil-levels of wealthy.

What happened:

• Samsung crossed the trillion-dollar mark on the back of soaring AI memory demand.

• Nvidia is already pushing memory makers toward 16-layer HBM (the ultra-dense, ultra-fast memory that feeds AI chips) for delivery as early as late this year.

• Nvidia is also working on something called “GPU-Direct Storage” for its next-gen Vera Rubin chips, plus a newer memory type called HBF (high-bandwidth flash) that offers roughly 10x the capacity of today’s HBM.

• Samsung, SK hynix, and Micron are pouring a combined $50+ billion into new memory factories.

• Those factories take 18 to 24 months to build, so relief for your wallet is still a 2027-or-later situation.

The shortage isn’t just pain. It’s also the single biggest wealth-creation event in the history of the chip business. The companies that make the picks and shovels (memory makers, packaging firms, the whole supply chain) are printing money. Samsung joining the trillion-dollar club is the loudest possible signal that the AI buildout is real and it’s nowhere near done (much to the chagrin of PC enthusiasts).

For the innovators reading this: the constraint is the opportunity (or more like, the problemtunity). Every dollar that memory eats is a dollar that rewards software which does more with less. Lean models, smart caching, efficient inference. The crunch isn’t going away soon, so the skill of squeezing value out of limited compute is about to be worth a whole lot.

The Weekly Whoops: the guy who wrote a book about AI lying… got caught lying with AI

The story: Steven Rosenbaum published a book this month called The Future of Truth: How AI Reshapes Reality. It’s a serious-minded warning about how AI distorts what’s real. There’s just one teensie-weensie, itsy-bitsy problem. The New York Times found that the book itself is sprinkled with more than half a dozen fake or misattributed quotes, apparently generated by the same kind of AI chatbots the book is warning you about. Yes, really.

What happened:

• One fabricated quote was pinned on tech journalist Kara Swisher, who told the Times she “never said that” and that the AI made her “sound like I have a stick up my butt.”

• Neuroscientist Lisa Feldman Barrett got quotes attributed to her that she said “don’t appear in the book and they are also wrong,” the kind of thing she “would never say.”

• Rosenbaum admitted the book had “a handful of improperly attributed or synthetic quotes” and opened an investigation into his own book (which is a sentence I enjoyed typing).

• He disclosed in the acknowledgments that he used ChatGPT and Claude during research and writing.

• His defense, more or less, was that if this serves as a warning about the risks of AI-assisted research, well, that’s the whole point of the book. Bold move, calling the fire your house caught a “fire safety demonstration.”

Here’s the actual lesson buried under the irony. AI is a phenomenal research assistant and a confident liar at the same time (kind of like that one intern, that one summer). The tools will hand you a perfectly phrased, perfectly formatted quote that the “source” never said, and it will look exactly as real as the true ones. There’s a name for it: hallucination. The model isn’t trying to deceive you. It’s pattern-matching what a quote should sound like, and sometimes the pattern wins over the facts.

So if you’re using AI to write anything that has your name on it (a report, an article, a book about truth, or even a weekly newsletter with charming humor that I won’t directly name), verify before it ships. Check every quote, every stat, and every citation against the real source. Do the fact-checking with your own two eyes. That should keep you off the wrong kind of New York Times news page.

What ties it all together

Last week was about distribution. This week was about everything that happens once the distribution land-grab is on: the Clone talent war heats up, the lawsuits get settled, the money gets… money-er, and the companies feeding the whole machine cross into rich Uncle Moneybags territory.

Step back and look at the board. Google is racing to put an agent in two billion pockets (I just checked my pockets for a tiny person). Anthropic is collecting the researchers who build the frontier. OpenAI cleared its biggest legal cloud and turned on the ad money. And Samsung crossed a line only a handful of companies in history ever have, purely because the world cannot get enough AI memory.

The thing worth holding onto is how fast all of this is compounding. Models, agents, talent, money, hardware: each one pulls the others forward. A year ago, any single one of these stories would have been the headline of the month. This week, it’s just business as usual, ‘round these here Weekly Think parts.

What I’m watching

• Whether Gemini Spark actually works as advertised when it hits Ultra subscribers next week, or whether it’s another impressive demo

• Where Karpathy’s pre-training team shows up first in Claude’s capabilities

• Whether Musk’s appeal goes anywhere, or whether this is really the end of the saga

• How people react to ads showing up inside ChatGPT (tolerance or revolt?)

• Whether any other memory maker chases Samsung into the trillion-dollar club this year

  

Thanks so much for reading this edition of The Weekly Think.

See you next week, matey!

Share

Subscribe now

When I was a kid, “AI” meant one thing and exactly one thing. It was the opposition.

The boss at the end of the level. The enemies between you and the boss. The obstacles, the rival racers, the CPU in NBA Jam who somehow knew you were going for that slam dunk. Whatever shape it took, the objective never changed. You were there to beat the AI, no matter how clever it was (if you got frustrated, you lowered the difficulty to “easy”😓).

Those were very cool afternoons. Firing up the SNES to see what tomfoolery would come out of the latest rental from Eastman Video (which later got bought out by Hollywood Video, a sentence/thought that makes my back hurt). A slice of pizza in one hand, a joypad in the other, and a single clear mission for the day.

That’s the thing I keep coming back to lately. The word “AI” was so simple back then. It was a thing you beat on the weekend, in a tiny bedroom that was way too small for 10 kids (sorry, mom & dad).

It has not stayed that simple. The word has worn three completely different costumes across one lifetime, and watching it change clothes explains a lot about why this particular moment feels the way it does.

When AI was the opponent

The first costume was the one I just described. AI lived on the other side of the screen, and your entire relationship with it was adversarial by design.

AI was NOT your friend. You didn’t ask it for help. You studied its patterns, learned where it was weak, and exploited that weakness until the words “GAME OVER” stopped showing up on your side of the screen.

I want to plant that idea here and let it marinate, because it matters much more than meets the eye (they’re robots in disguise!… sorry, that’s a Transformers reference. Better save that for a later post.) The very first relationship most of us ever had with the term was a fight. We learned the word in a setting where defeating it was the whole point.

When AI became the sentient machine

The second costume came courtesy of Hollywood, and it grew up right alongside me.

Somewhere along the way, AI stopped meaning M. Bison from Street Fighter II and started meaning the machine that has an awakening. Sometimes the machine was kind and worked for the good of humankind. Sometimes the machine looked at humankind, did some quick math, and decided, “yeah, gonna get rid of these fleshbags real quick”. You already know the movies.

I’d rather point at the stranger one. There’s a Spielberg film from 2001 called, quite fittingly, A.I. It wasn’t the friendly-robot story and it wasn’t the destroy-all-humans story. It was the uncomfortable middle, a story about a machine that mostly just wanted to be loved and a world that didn’t know what to do with it. Nobody in that movie quite knew which camp to be in.

What a wild time. For this entire stretch, AI was not something you used. It was something you imagined, and mostly something you were a little afraid of.

When AI showed up on your actual desk

Then the third costume arrived, and it’s the one we’re all still adjusting to. AI stopped being imaginary. It became a tool sitting right there on the desk, doing actual work.

It helps researchers move faster through the kind of deep work that used to take months. It shows up in healthcare. It proofreads the email you were too tired to read twice. It sits in the room while you write and offers you a second set of eyes that never gets bored.

I’ll say my own piece plainly, because I’ve earned the right to an opinion here. I’ve worked as a multimedia creative for more than twenty years, and I find AI genuinely useful in my own writing and creative process. It’s one of the most capable creative tools I’ve put my hands on. I’m not embarrassed about that and I’m not going to pretend otherwise.

And in the very same breath, I cannot stand the slop.

You know the slop. The thoughtless, obviously-generated content, churned out by people who clearly didn’t read it before they posted it, recycled endlessly to the benefit of just about no one. It’s lazy and it’s just about everywhere (especially on your parent’s Facebook feed). And it gives the whole technology an off-odor that the technology doesn’t entirely deserve.

Here’s the part that the loud people on both sides have trouble with. I think the tool is remarkable and I think a lot of what it produces is garbage, and I hold both of those opinions without straining. One doesn’t cancel the other. I don’t have to talk myself down from the praise to make room for the criticism, or soften the criticism to keep the praise.

Why we still want to beat it

So here’s the strange place we’ve all ended up at.

One group of people loves this technology and wants to use it for everything. Another group wants to burn it to the ground in all of its forms. The conversation between them is less of a conversation and more of a standoff, and most of us are standing in the middle wishing everyone would lower their voices (I work at 6:00 a.m., thank you).

And I think, in a way, we’re right back in those old video game afternoons. A lot of people, deep down, just want to beat the AI.

That reflex makes a strange kind of sense once you remember where it came from. We met this thing as an opponent. The first definition any of us ever learned was the one where the entire point was to defeat it. So when a genuinely useful tool shows up on the desk decades later, some part of the old wiring still reads it as the boss at the end of the level. Or even a scary thing that threatens to take over the planet. The frame never got updated. The instinct to win against it is a leftover from a version of AI that either doesn’t exist anymore or was made up by some dudes in a writer’s room.

But the thing on your desk isn’t a boss fight. It’s not an overlord. It never picked a side. It’s just sitting there, capable of real help and real garbage depending entirely on who’s holding the controller.

The position nobody will let you hold in peace

The useful stance, the one I keep landing on, is the balanced one. And it happens to be the one neither camp will let you stand on without yelling at you.

You’re allowed to praise it and criticize it in the same sentence. You’re allowed to find it indispensable for your work on Monday and roll your eyes at a slop article on Tuesday. Knowing the nuance of where it actually helps and where it actually hurts is the entire practical skill, and it’s a skill, not a betrayal of one team or the other.

Whether you use it at all is a decision that lands on you, and honestly, it should. Not on the camp that would shun you for touching it, and not on the camp that would high-five you for posting something you didn’t write. It’s all on you.

Maybe one day the two sides drift toward the middle, and AI gets seen the way we eventually see every tool, as a thing that’s neither good nor evil, just useful in some hands and wasteful in others. I’d like to be around for that.

I miss the joypads

Some afternoons I do miss the old definition. The one where AI just meant the latest RPG to outwit, with an SNES controller in one hand and a soda within arm’s reach of the other, treating Mountain Dew like it was the water of life.

Although, on reflection, maybe I’ll leave that last Mountain Dew-fueled part in the past. My 40+ year old body tummy aches, just thinking about it.

The controller is gone either way. The thing I used to spend weekends trying to beat now sits quietly on my desk, waiting to help me write the very sentence you’re reading. Though, I guess a part of my noggin is still saying “beat the AI!”. Maybe that’s the actual work now, for all of us. Not beating it, and not worshipping it. Just finally seeing it for what it is.

Subscribe now

AI news seeker, come on down! Last week, I called what these labs are turning into a sort of FrAInkenstein (note to self: gotta trademark that bad boy). This week kind of proved the point. Nobody shipped a flashy new frontier model. What they shipped instead was distribution: into law firms, small businesses, Microsoft Office, and Beijing. Oh my!

The pattern isn’t hard to recognize. The race has shifted to distribution: who can plant their flag inside the most workflows before anyone else does. And the labs are basically copying Palantir’s old playbook now (more on that in a sec).

Here’s what’s covered today:

• Anthropic launched Claude for Legal AND Claude for Small Business in 48 hours

• OpenAI started a $4B consulting firm and called it the Deployment Company

• Google previewed its “Gemini Intelligence” and a new laptop category called Googlebook

• Jensen Huang got a last-minute call from Trump and ended up in Beijing

• The memory shortage went from “concerning” to “no affordable laptop for you!”

Anthropic went vertical: Legal on Tuesday, Small Business on Wednesday

The story: On May 12, Anthropic launched Claude for Legal: 20+ legal MCP connectors, 12 practice-area plugins, integrations with Westlaw, Thomson Reuters, Harvey, Everlaw, the works. Less than 24 hours later, on May 13, they shipped Claude for Small Business, a one-click install that wires Claude into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365.

What happened:

• Freshfields, one of the world’s biggest law firms, has already deployed Claude to thousands of users.

• Legal became the #1 power-user job function inside Claude Cowork, with 3x the usage of any other vertical.

• Daniela Amodei said small businesses “make up nearly half the American economy” and have never had enterprise-grade tools.

• Anthropic also bundled a free AI fluency course (14 lectures, an hour of video) for the small business launch.

This is what going vertical looks like, friendos. A year ago, Anthropic was a chatbot. Now it’s an orchestration layer that sits between you and your billable hours, or between you and your unpaid invoices. The thing that should jump out at you, though, is the pacing. Two completely different verticals, two days apart, both shipped with real partner integrations already in place. Good gravy! Companies that take eighteen months to launch a single integration should be taking notes.

If you’re a small business owner reading this and you haven’t dug into Claude yet, this is the week to start. An AI course priced at free 99 is a great place to start.

OpenAI launched its own consulting firm (and immediately bought one)

The story: On May 11, OpenAI announced the OpenAI Deployment Company (also known as DeployCo). It’s a $4 billion subsidiary backed by 19 outside investors, with the very specific goal of, quite literally, putting OpenAI engineers inside Fortune 500 buildings. The same day, OpenAI announced it’s acquiring Tomoro, a London/Edinburgh-based AI consulting firm, to staff the whole ding darn thing on day one.

What happened:

• Tomoro brings approximately 150 Forward Deployed Engineers from day one (Palantir invented this role, and it apparently prints money).

• DeployCo launched with TPG as lead investor, plus Bain Capital, Brookfield, Goldman Sachs, SoftBank, and Warburg Pincus on the cap table.

• Three of the largest consulting firms in the world (Bain, Capgemini, McKinsey) also wrote checks.

• Tomoro’s existing client list reads like a fun trivia card: Tesco, Virgin Atlantic, Red Bull, Mattel, the NBA, Supercell.

• Axios reportedly values the unit at $14 billion.

You catch what just happened? OpenAI just got into the consulting business. They’re sending their own engineers into big companies (think Tesco, Virgin Atlantic, the NBA) to build custom AI tools on-site. The money isn’t in having the smartest model anymore. It’s in the hands-on work of making AI actually useful. Anthropic has been doing this with one law firm at a time. OpenAI just hired 150 engineers to do it with hundreds of companies at once.

For software builders, this is the kind of move you should be taking notes on. The deployment problem is so big that one of the biggest AI labs in the world just spun up a 150-person team specifically to solve it for other people. If you can be the person who walks into a company and makes their AI actually work, you’re going to have more work than you can handle for the next few years.

Google previewed Gemini Intelligence, a new laptop, and basically told Microsoft “we’ll take it from here”

The story: On May 12, Google held its Android Show: I/O Edition, a warm-up event a week before the main Google I/O keynote on May 19. They previewed Gemini Intelligence (a suite of agentic features baked into Android), Gemini in Chrome with “auto browse” (the browser, browsing itself, which is very meta if you ask me), and Googlebook, a new laptop category built for Gemini from the ground up.

What happened:

• Googlebooks will be made by Acer, ASUS, Dell, HP, and Lenovo, shipping this fall.

• The new OS underneath them, called Aluminium OS, is essentially Google merging Android and ChromeOS into one stack.

• A feature called Magic Pointer lets you wiggle your cursor at anything on screen, and Gemini suggests what to do with it (one writer compared it to Clippy “but potentially 1,000 times more annoying,” which, like, justice for Clippy bro✊🏽📎).

• Gboard is getting Rambler dictation cleanup, which scrubs up the “uhh… so basically” mess in your voice notes.

• Android 17 is also adding Pause Point, which interrupts you before you can open the doomscrolling app of your choice.

Microsoft tried this with Copilot+ PCs in 2024 and it didn’t exactly catch fire. Engadget had a great line on this: Microsoft “started stripping Copilot features out of Windows apps after users revolted.” So Google is walking into a category that the much bigger software giant already faceplanted in. Confidence is a heck of a placebo.

Two things you should be watching: first, whether Magic Pointer is genuinely useful or just genuinely annoying (try wiggling your finger at grocery store items with people around; see what happens). And second, whether Gemini in Chrome browsing-on-your-behalf is the moment agentic AI stops being a demo and starts being a habit. Speaking of which, I/O is next Tuesday. You better believe I’ll have thoughts.

Jensen Huang got a last-minute call from Trump and ended up on Air Force One

The story: On Monday, May 12, a list of CEOs traveling with Trump to meet Xi Jinping in Beijing went out to the press. Apple’s Tim Cook? Yes. Tesla’s Elon Musk? Yes. Qualcomm’s Cristiano Amon? Yes. Nvidia’s Jensen Huang? Notably… no. The CNBC headline was basically “oof.” Cue the actually-watching-his-own-funeral SpongeBob meme.

What happened:

• After CNBC and others ran the story of Huang’s absence, Trump personally called him.

• Huang flew to Alaska to board Air Force One mid-trip (which is a sentence I’d like to revisit when I’m old).

• Trump posted that opening up China for U.S. businesses would be his “first request” to Xi.

• Huang’s Nvidia made up the entirety of Anthropic’s SpaceX Colossus 1 lease last week (220,000 GPUs of it, anyway).

• Hassam Nasir at Tom’s Hardware noted Nvidia’s most advanced chips have faced tighter U.S. export restrictions on China for four years running.

Imagine being so necessary to global AI infrastructure that the President of the United States personally rerouted you onto a presidential aircraft in Alaska (in a huge “my bad, dawg” situation). Most of us would feel like we just got picked for the kickball team. Jensen probably felt like that was just another Tuesday. There’s a real story here about how the most consequential (and probably polarizing) CEO in tech right now is the one making the picks and shovels for everybody else’s gold rush. It’s reality: without Nvidia, this whole AI party probably wouldn’t be so crazy.

The compute supply situation is going to keep shaping what models you can use, where you can run them, and what they cost. Speaking of compute…

The memory shortage is now everyone’s problem (yes, including yours)

The story: Two reports this week made the memory crunch impossible to ignore. Insurance-Edge published a deep dive on May 13 showing that DRAM contract prices surged ~90% in Q1 2026, with order lead times now running past 40 weeks. Cisco, also on May 13, posted a 15% stock pop on earnings, and announced 4,000 layoffs starting May 14 to refocus on AI demand. Memory and AI are eating everything else.

What happened:

• Counterpoint Technology Market Research clocked memory prices up ~90% in Q1 alone.

• IDC projects 2026 DRAM supply growth at just 16% year-over-year, well below historical norms.

• Samsung’s memory chief Kim Jaejune warned shortages will persist through “at least 2027.”

• Some hyperscalers have already locked in HBM supply through 2027.

• Cisco’s revenue grew 12% on AI demand and they’re still cutting 4,000 jobs to chase it harder.

Here’s why this matters even if you never touch a data center (because, man, I’ve never even SEEN one). When the three companies that make 90%+ of the world’s memory (Samsung, SK hynix, Micron) decide to send their best wafers to Nvidia instead of Dell, the cost of every laptop, every phone, every PlayStation (gamers: “noooooo”), and every smart fridge (produce in the fresh drawer: “noooooo”) goes up. Lenovo, Dell, HP, Acer, and ASUS have all warned of 15-20% price hikes (and look, I love hiking but… oh, not that kind of hike…). Your next PC upgrade is being indirectly funded by Anthropic and OpenAI buying GPUs. (Funny how that works, no? Actually, it’s not funny.)

The flip side is the opportunity. If you’re a software builder or a small business owner, the memory crunch is creating a giant moat around anyone who can squeeze more value out of less compute. Models that run efficiently on local hardware just got a lot more interesting, and open-weights players did too. The companies betting big on running everything in the cloud are going to find out how expensive “the cloud” really is.

What ties it all together

Last week, I said the “AI lab” category was collapsing in real time. This week was the first week where you could actually see what’s replacing it.

You’ve got Anthropic building vertical-specific products (Legal, Small Business, Financial Services) the way a normal enterprise software company would. You’ve got OpenAI standing up a 150-engineer consulting arm to sit inside your office (gonna one heck of a stuffy office; hope there’s enough coffee). You’ve got Google merging two operating systems and shipping a laptop. And underneath all of it, you’ve got Jensen Huang on a plane to Beijing and a memory shortage that’s reshaping the price of every device you own. Whew.

The big tell of the week is what the labs are now selling. It’s integration. They’re embedding inside legal workflows, inside small business stacks, inside Microsoft Office, inside laptops, inside chip-supply diplomacy. The companies (and the people) who figure out how to ride that wave, by getting deployed inside workflows that actually matter, are the ones who win the next 12 months.

What I’m watching

• Whether Google’s I/O keynote on May 19 lands Gemini 4 or settles for an incremental bump

• How fast OpenAI’s DeployCo eats into Accenture/Deloitte’s AI services revenue (early signs not great for the consultancies; Accenture dropped 3% on the announcement)

• Whether Anthropic adds a vertical-specific Claude every week now (Claude for Healthcare? Claude for Real Estate? Claude for Farming??)

• Whether Trump’s China trip produces any actual chip-export movement

• DRAM spot prices, because the whole industry might be about to find out it built on sand (and not rock, like a smart person)

  

Thanks so much for reading this edition of The Weekly Think.

See you next week!

Share

Subscribe now

Let’s say you’ve been pitching ghostwriting clients for what feels like months now. You’ve found prospects on LinkedIn, on X, in newsletter archives, and in those “founders to follow” lists everyone shares around like a chain letter. You’ve crafted thoughtful cold emails. You’ve sent free outlines. You’ve hopped on a discovery call or two.

And the response, mostly, has been a polite “we’ll be in touch.” Or, more commonly, the kind of silence that makes you wonder if Gmail is even still delivering your messages or if it’s just keeping them as a souvenir.

If this has been your rinse and repeat for a while now, it’s easy to see why you might be losing steam. You might even be wondering if you should shelve the whole ghostwriting thing and go back to whatever you were doing before. And worst of all, that little voice in the back of your head might be whispering the worst possible thing it can whisper.

Maybe my writing just isn’t good enough.

Let’s pump the brakes on that one.

You are not the problem

I want to say that as plainly as possible, because the longer you sit with the other story, the more convincing it gets.

If you are out here pitching clients, writing free outlines, taking the calls, and putting yourself in the arena (which, by the way, is a really uncomfortable arena and not enough people will tell you that), you are most definitely a skilled writer with the drive to do something real with the craft. You wouldn’t be pitching at all if you weren’t. People who aren’t serious writers do not voluntarily spend their free time writing a 1,200-word outline for a stranger on LinkedIn. That’s simply not a thing, o friend of mine.

Silence from a prospect (or even a flat “no, thanks”) is not the same as “you aren’t good enough.” It usually just means the offer didn’t land at the right moment, for the right person, in the right way.

Here’s a comparison that might help. Picture yourself walking into a coffee shop on a Tuesday morning, holding a tray of perfectly made croissants you baked yourself, and offering one to the first person you see. They wave you off. Are the croissants bad? No, the croissants are great. The person already ate breakfast. The person is on a call. The person is gluten-free. The person was raised in a household where you do not accept baked goods from strangers in coffee shops (though, to be fair… I would have asked if I could take five… but that’s another story for another day).

The croissants are fine. They’re FIRE.

Your pitches are croissants. And they are not ON fire.

Why this stretch feels worse than it is

There’s something weird about cold outreach that nobody really warns you about when you start. In most other parts of life, effort and outcome are at least loosely related. You study, you do better on the test. You practice the song, the song gets sharper. You go to the gym, eventually a shoulder appears where there wasn’t one before.

Cold outreach does not work that way, and the brain handles it badly.

You can do everything right and hear nothing. You can do everything right for months and hear nothing. And because the only feedback you’re getting is an empty inbox, the brain starts filling in its own explanations for why. The brain, as a general rule, is not great at this. The brain will reach for the most personal, most catastrophic explanation available, because the brain has been watching too many movies and thinks every plot needs a third-act crisis.

The actual answer is almost always much more boring. Your prospect was in a meeting. Your prospect was on vacation. Your prospect opened the email on their phone, meant to respond later, and then their toddler threw a yogurt and that was the end of that. None of those things have anything to do with you.

What to do instead of spiraling

The first thing is to take the question seriously enough to set it down. Not bury it or pretend you never asked it. Set it down, the way you’d set down a coffee mug you’re not ready to put in the dishwasher yet. It’s there; you’re just not drinking it right now.

The second is to remember why you started this. You picked ghostwriting because you wanted to take ownership of your work and your time. You niched down because you wanted to be the obvious choice for a specific person. You wrote those free outlines because you actually wanted to help. None of that has changed because the inbox is quiet this week.

The third is the part of the advice that everybody needs. Lean on your people. The friends who will pick up the phone. The family member who always asks how it’s going. The writing group. The one weird friend who replies to your voice notes with their own three-minute voice notes (best believe everyone has this friend and they are a treasure). Encouragement from people who actually know you carries way further than any motivational post on the internet ever will.

And the fourth is permission to step away from the pitching grind every now and then without feeling like you’re slacking. Take the walk. Take the afternoon. Watch a movie to relax (or maybe even laugh until you cry). Give yourself a little of that “me time” so the work doesn’t start to feel like a punishment for the crime of being a writer.

The part where I tell you to keep going

Your persistence is going to pay off. That’s a sentence I’m willing to put my name on, because the writers who keep refining and showing up are the ones who eventually land the right fit. The client you’ve been hoping for is, in all likelihood, much closer than the inbox is making it feel.

And because you’ve taken the time to niche down, sharpen your offer, and actually understand the people you’re reaching out to, when “yes” finally comes, it’s going to be right on time. The reason matters here. You aren’t trying to land any client. You’re trying to land the client who is a fit for the specific thing you’re good at. Those clients exist. You’re already doing the work to find them.

The ghostwriting hunt is hard when it isn’t producing results yet. Most worthwhile things are. The next email you send could honestly be the one that changes the whole picture, which is a thing I say not because it’s a motivational fortune cookie but because that is mathematically how this kind of work tends to go. The “yes” doesn’t announce itself in advance. It just shows up one Tuesday while you’re making coffee.

So go send the next one. Refill your croissant tray. Walk back into the coffee shop.

You’re a lot closer than the inbox is letting on.

And if you need someone to hear you out, my DMs are always open. And free. Like unexpected croissants!

Good day, weary AI news traveler! A lot happened in 24 hours, and y’all, none of it was about new models (though a fresh new face in this space sure would be nice, no?).

Anthropic’s annual run rate hit $30 billion and passed OpenAI for the first time. They leased every GPU at SpaceX’s Colossus 1 facility. Then the Trump administration cut a federal AI testing deal with Microsoft, Google, and xAI, and didn't invite Anthropic to the table.

You know what this means? The good ol’ fashioned “AI lab” category is collapsing in real time. What’s replacing it is something… welp, altogether different: part defense contractor, part cloud provider, part financial services firm, part aerospace company. A sort of FrAInkenstein, if you will. I should trademark that.

Subscribe now

Here’s what’s covered today:

• Anthropic’s revenue blows past OpenAI

• The Anthropic-SpaceX compute deal nobody saw coming

• Anthropic gets locked out of the federal AI deal

• A court just made AI ad platforms liable under securities law

• Apple opens iOS 27 to third-party AI

Anthropic’s revenue blew past OpenAI

The story: Anthropic’s annualized revenue hit $30 billion this week. OpenAI is at $24 billion. Six months ago that ratio was, as Fresh Prince once said, “flipped, turned upside down”. Dario Amodei told a room of bankers they’d projected 10x growth and got 80x instead. Mr. Monopoly was not available for comment.

What happened:

• Anthropic shipped 10 preconfigured Claude agents Tuesday for investment banks, asset managers, and insurers.

• Claude Opus 4.7 dropped alongside them, tuned for financial work.

• A $1.5B joint venture with Blackstone, Goldman, and Hellman & Friedman launched the day before to embed Claude into mid-market financial firms.

• Jamie Dimon shared a stage with Amodei at the New York briefing.

This is what happens when “best at coding” turns into “best at coding workflows that big companies pay billions for.” Anthropic is winning enterprise. Enterprise is the only customer base where the unit economics actually work for a frontier lab. Wall Street can pay frontier-AI prices forever, and Anthropic just made themselves the operating layer for it.

The Anthropic-SpaceX deal nobody saw coming

The story: Anthropic announced a deal Wednesday to lease all of Colossus 1, Elon Musk’s 300+ MW facility in Memphis. Yeah man, the same Elon who spent the last year calling Anthropic “doomed to become misanthropic.” To quote one Michael Scott of The Office, “Well, well, well, how the turntables…”

What happened:

• 220,000+ Nvidia GPUs come online for Claude within the month.

• Claude Code’s 5-hour usage caps are doubling across paid tiers.

• Anthropic also “expressed interest” in working with SpaceX on multiple gigawatts of compute capacity in space (yes, in space).

• Musk posted on X: “no one set off my evil detector.”

When the world’s most committed troll calls a truce with the company he’s been dragging for eighteen months, that’s quite the market signal, my dudes. The compute crunch is real enough that even Musk is choosing money over grudges.

Anthropic got locked out of the federal AI deal

The story: While Anthropic was signing with SpaceX and Goldman, the Trump administration cut a separate testing deal with Microsoft, Google, and xAI. Anthropic wasn’t invited to the party; they were basically treated like Cady in Mean Girls. The DoD’s classified-network deal last week with 10 AI providers? Same story.

What happened:

• The exclusion ties to Claude Mythos, Anthropic’s restricted frontier model.

• Mythos found thousands of zero-day vulnerabilities (including a 27-year-old bug in OpenBSD) during internal testing.

• Federal policy has shifted from “release with guardrails” to “don’t release at all” for offensive-capable models.

• Mythos access is currently limited to 12 partner orgs under Project Glasswing (AWS, Apple, Cisco, Google, JPMorgan, Microsoft, etc.).

The lab that’s leading on revenue (sweet!) and capability (nice!) is also the one most isolated from federal procurement (*record scratch*). That’s a strange position to be in. Two readings: either Anthropic’s safety posture is now strict enough that they’re harder to contract with, or the government doesn’t want the lab whose model just made it rewrite tech policy.

A court just made AI ad platforms liable for what their AI says

The story: *Cue the theme from Dragnet* A federal court in the Northern District of California ruled that when a platform’s AI exercises “ultimate authority” over assembled ad content, the platform itself can be considered a maker of fraudulent statements under Rule 10b-5 securities law. Dun-dun-dun-dun-dunnnnnnn.

What happened:

• Meta, Alphabet, Snap, TikTok, and X all run generative AI in their ad pipelines.

• Every one of them is now legally on the hook for what their AI puts in ads, and none of them have a plan for it.

• Separately, U.S. lawyers are warning clients that AI chatbot conversations may be admissible in court.

• None of those Claude or ChatGPT conversations have legal privilege.

AI agents are entering every layer of commerce, law, and corporate operations faster than the legal frameworks designed to govern them. The next 18 months are going to produce a parade of “wait, that’s legal?” moments. If you build software, you should be seeing dollar signs, like you’re an old timey cartoon. There’s a whole new market for tools that help companies stay legal as AI takes over their workflows.

If you’re an AI user? Be careful what you type. Your ChatGPT history isn’t private, and lawyers are already figuring out how to subpoena it.

Apple opened iOS 27 to third-party AI

The story: Apple announced that iOS 27 will let users pick which AI service powers their device-level features. It’s Apple saying “pick a card” at the OS level.

What happened:

• Anthropic, OpenAI, Google, and likely a Chinese open-weights player are all in the running.

• Apple has spent 18 months trying to ship their own AI strategy and produced approximately nothing usable.

• This pairs with Salesforce’s headless announcement last week, which exposed their full SaaS platform to AI agents.

• Two billion devices just became default-level real estate for whichever lab the user picks.

Apple basically gave up trying to build their own AI. Instead, they’re letting Anthropic, OpenAI, and Google compete on the iPhone, and Apple takes a cut from whoever the user picks. They’re not worried about winning. They’re “the house”, and in the parlance of poker, they always win.

What ties it all together

This week made one thing clear: what we used to call an “AI lab” doesn’t really exist anymore. A year ago, Anthropic and OpenAI were research companies that sold a chatbot. Now they’re leasing data centers, building compute in space, running Wall Street agents, and rewriting federal tech policy. But, they aren’t making julienne fries… yet.

At any rate, these aren’t AI labs anymore. They’re something else, and honestly, we don’t have a name for it yet. The category is being invented in real time, which means there’s never been a better time to break out the popcorn and see what transpires.

What I’m watching

• Whether the federal exclusion of Anthropic becomes a bigger story

• Whether any other lab delves into orbital compute

• Which AI services Apple names for the iOS 27 launch list

Thanks so much for reading this edition of The Weekly Think.

See you next week!

Share

Subscribe now

I’ve been experimenting with short-form writing for the past few months and noticed something I didn’t expect: tighter constraints produce better results. And by “better” I don’t mean more polished or more professional. I mean more people actually read the thing.

That sounds obvious when you say it out loud. Shorter writing gets read more. Groundbreaking stuff. But the way I got there was a little unconventional, and the tools I used along the way surprised me. So I figured I’d share the whole process.

The rules

Before I started writing, I built a constraint set. Every piece I publish has to follow it:

95 words or fewer. No em dashes. No bold text. No emojis. Written in second person. Staccato rhythm. Every post ends with a reader-facing question.

That probably sounds restrictive. It is. That’s the point.

Social media is full of long, rambling posts that lose people by the third paragraph. I wanted mine to feel like a conversation at a coffee shop where you’ve only got two minutes before one of you has to leave. Short enough that someone scrolling at 7am actually finishes reading. Tight enough that every word earns its spot.

The 95-word cap alone changed how I think about writing. When you’ve only got 95 words to work with, filler evaporates. “I think it’s really important to consider the fact that...” becomes a dead sentence on arrival. You learn to get to the thing.

Where Claude fits in

I use Claude (Anthropic’s AI, for the uninitiated) as a drafting partner. And I want to be specific about what that means, because “I use AI to write” can sound like I’m hitting a button and collecting output. The actual workflow has more friction than that.

It starts with an idea. Usually something I’ve been chewing on: a news story in AI, an observation about how founders talk about their work, a pattern I keep noticing online. I bring that seed to Claude along with my constraint set and a voice document (basically a profile of how I write, my humor, my quirks, the things I’d never say).

Claude produces a draft. I read it. And honestly? The first draft is almost never ready to post. It’s close in the way that a cake with the right ingredients but the wrong proportions is close. All the pieces are there. The balance is off.

So I edit. I cut. I rearrange. Sometimes I throw the draft out and ask for a different angle entirely. The voice document helps Claude get in the neighborhood, but I’m the one who parks the car.

(That metaphor got away from me a little, but you know what I mean.)

What I’ve noticed

Two things surprised me early on.

First: certain styles of writing attract more readers. I’ve been publishing these short pieces on LinkedIn, and the posts that followed my constraints consistently outperformed the ones where I got lazy and let a post run long or broke my own rules. Turns out people are grateful when you respect their time.

Second: the writing I tuned specifically for AI executives landed well with casual readers too. I expected the opposite. I figured posts aimed at CTOs and founders would feel niche or inaccessible. They didn’t. The specificity made them more interesting. A post about how a Series B founder should think about their public narrative reads differently than generic profile tips. Even people outside that world found something useful in it.

Why I’m writing this

The short version: I’m a ghostwriter, and AI is baked into how I work. This is where I take that process apart and show you how it works.

I’m going to document the tools, the workflows, the experiments. What works, what falls flat, what I’m still figuring out. If you’re curious about AI, writing, or the weird overlap between the two, you’re in the right place.