I've been away from this list for a while. Seven weeks, if we're counting, which we are, because I went and looked it up and winced a little.
Summer got busy. That's the whole explanation, and it's not a very good one. The world didn't wait around for me to get un-busy, so this isn't a week's worth of news. It's the back half of the summer in one sitting, and I had to be more ruthless than usual to cut it down to ten, because there was genuinely more real news in these seven weeks than in some full years I've written about.
Same format as always. What actually happened first, the names and the numbers, then a little of what I make of it. Let's get into it.
AI: The Money and the Reckoning
1. Anthropic and OpenAI both filed to go public, and the numbers are hard to hold in your head
Both of the biggest AI labs are heading for the public markets at the same time. Anthropic confidentially filed its S-1 with the SEC on June 1 and OpenAI followed a week later on June 8, and through August both were out meeting investors ahead of share sales that could launch as soon as September or early October. Anthropic's last private valuation was around $965 billion, set by a $65 billion round that closed in late May. OpenAI ran a $7 billion buyback of employee shares in August, priced at a $852 billion valuation. You will also see $920 billion quoted for OpenAI; that comes from a different moment, and the buyback price is the one with a transaction behind it. Goldman Sachs and Morgan Stanley are running the books on both deals, and each one is expected to raise at least $60 billion.
I keep re-reading those figures to make sure I've got the right number of zeros. What it means practically is that regular people, through index funds and retirement accounts, are about to own a piece of this whether they ever decided to or not. I'm not saying don't buy. I'm saying know what you're buying, because "AI company" is doing an awful lot of work inside those valuations, and the public market is where a private story finally has to show receipts every ninety days.
2. Anthropic's $1.5 billion settlement with authors got final approval
On July 20, Judge Araceli Martinez-Olguin in the Northern District of California granted final approval to the settlement in Bartz v. Anthropic, the case over books used without permission to train its models. It's the largest copyright settlement on record and the first major US AI copyright case to actually settle. The settlement pays $3,000 per work across roughly 482,000 eligible works. The judge trimmed the attorneys' fees to about $101.6 million, down from the $187.5 million class counsel asked for, and around 350 authors opted out to chase their own claims instead.
So now there's a price on the thing everybody had been arguing about. For years the training-data question was basically philosophical. Is it fair use, is it theft, is it some third thing we don't have a word for yet. A number ends that conversation in a hurry. Three thousand dollars a book isn't justice exactly, but it establishes that the material these models learned from was never free, and every lab building the next one carries that on the balance sheet now.
3. The AI jobs story stopped being a forecast and became a count
For a couple of years this was mostly projection. It isn't anymore. Through July, Challenger, Gray & Christmas counted 112,713 job cuts this year where the employer named AI as the reason, about 24 percent of every cut they tracked. That has made AI the single most-cited reason five months running. The cuts cluster in customer service, data operations, entry-level software, and finance back offices. The sharper number is on the hiring side: new graduates made up just 7 percent of big tech hires in 2024, down more than half from where it sat in 2019.
The shape of this keeps being the real story. It isn't a wave of people getting walked out of buildings. It's doors that quietly stop opening. Layoffs are loud and somebody counts them. A job that never gets posted doesn't show up in any tally at all, and that's where most of the damage is landing. The advice that worked for my generation, get in somewhere and learn on the job, quietly assumes there's an in.
Medicine Had a Real Summer
4. The FDA approved a first-of-its-kind pancreatic cancer drug that nearly doubled survival
On August 26, the FDA approved Rasonque (daraxonrasib), from Revolution Medicines, for adults with metastatic pancreatic adenocarcinoma who've had at least one prior treatment. It's the first broad RAS-targeted medicine approved for the disease. In the RASolute 302 trial, 500 patients were split evenly between the new drug and the chemotherapy that's been standard. Median overall survival came in at 13.2 months against 6.7. Progression-free survival was 7.2 months against 3.6, the response rate was 30 percent against 11, and the risk of death dropped 60 percent. The Washington Post called it a landmark approval.
Pancreatic cancer is the one where the diagnosis has always worked more like a sentence. Thirteen months instead of seven sounds modest written out that way. But if you've ever sat with a family right after that particular diagnosis, you know what another half a year actually means, and it isn't measured the way a statistician measures it. The honest asterisk is that this extends life rather than curing anything. Still. RAS mutations got called undruggable for forty years. Somebody just drugged them.
5. A blood test predicted cognitive decline a decade out, and helped regular doctors catch it
The Alzheimer's Association International Conference ran July 16 to 21 in London, with more than 11,000 attendees and over 7,800 scientific submissions. Two findings stood out to me. In a study of nearly 2,700 cognitively healthy older adults, the people with very high levels of a blood marker called p-tau217 had an estimated 78 percent chance of developing cognitive impairment within ten years, and about a 33 percent chance within five. And in a Swedish study covering more than 1,300 patients and 165 physicians, family doctors who could see blood-test results diagnosed Alzheimer's correctly 93 percent of the time, against 94 percent for specialists. Without the test, primary care was at 65 percent and specialists at 76. The results changed the diagnosis or the care plan for more than half the patients in the study.
This is the one I'll admit I take personally, because dementia isn't an abstraction in my line of work. The second finding is the one that changes lives. For most families the road to a diagnosis runs through a regular doctor's office and then a long wait for a specialist who's booked months out. Closing a 65-to-93 gap with a blood draw means people find out sooner and closer to home. The first finding is harder to sit with. I genuinely don't know whether I'd want my own number.
The Machines Got Hands
6. The FDA cleared a robot that draws your blood
On August 19, the FDA authorized Aletta, made by a company called Vitestro, the first standalone device that can draw blood from a patient's arm with no hands-on human involvement. It uses near-infrared light and Doppler ultrasound to find a vein and tell it apart from an artery, then applies the tourniquet, preps the skin, inserts and disposes of the needle, swaps the collection tubes, and puts the bandage on. It cleared the FDA's De Novo pathway for adults in outpatient settings, supervised by someone trained in phlebotomy, and one person can oversee three machines at once. In the clinical testing the FDA reviewed, its success rate matched or beat trained human phlebotomists.
Most talk about automation stays abstract right up until it's holding a needle. This one's aimed at a genuine shortage, because the country hasn't had enough phlebotomists in years, so it's useful. It's also the clearest picture I've seen of how this actually goes. Not one robot replacing one worker, but one trained person running three stations. The job doesn't vanish. It gets leveraged, and then you need fewer of them.
7. A humanoid robot ran the 100 meters faster than Usain Bolt
At the second World Humanoid Robot Games in Beijing, August 22 to 26, 666 teams entered 2,056 robots across 51 events. A robot called Tiangong Ultra ran the 100 meters in 8.86 seconds in the semifinal, after posting a 9.39 on opening day. Usain Bolt's human world record is 9.58. Organizers said Tiangong machines also went past human world records in the 400 meters at 38.15 seconds, the 1,500 at 2:21.64, and the high jump at 2.88 meters. At the first edition of these games last year, the winning 100-meter time was 21.50 seconds.
Read that last line again, because it's the whole story. Twenty-one and a half seconds down to under nine, in one year. I want to be careful with it, since these are organizer-reported times on a track built for the occasion, and a machine that sprints in a straight line still can't fold your laundry. But the rate of improvement is the part I can't wave off. The honest question is what that number looks like next August.
Money and Home
8. The S&P 500 set records while inflation stayed stubborn
In mid-August the S&P 500 hit an all-time intraday high of 7,814.88 and set a closing record of 7,798.99. The Nasdaq finished at 26,803.03, and the Russell 2000, the small-cap index, set a record of its own at 3,067. The rally came off inflation data that landed cooler than economists expected, with both CPI and PPI below forecast, helped along by Brent crude falling 2.1 percent to $87.07 a barrel. The catch is that inflation is still running above 3 percent against the Fed's 2 percent target, and traders cut the odds of a September rate hike from around 50 percent down to 30.
What I like about this stretch is that the small caps came along for it. When only the giant tech names go up, the market is telling you a story about a handful of companies. When the Russell sets a record too, it's saying something about the actual economy underneath. I'd still hold the champagne, because "less bad" inflation above 3 percent isn't the same thing as solved. But a broad record beats a narrow one.
9. Medicaid work requirements are coming, and a bill would finally put assisted living on the books
Two things moved on Medicaid this summer. The work requirements from the 2025 reconciliation law got an interim final rule in June that took effect July 31, giving states their implementation guidance. Enrollees will need 80 hours a month of work, school, job training, or community service, and every state that expanded Medicaid, 41 of them plus DC, has to be running the program by January 1, 2027. Separately, Representative Max Miller of Ohio introduced a bill that would require state Medicaid programs to cover assisted living services for eligible low-income older adults, so long as the cost doesn't run higher than care in a hospital or nursing facility. Today Medicaid only reaches assisted living through state waivers that cover services but not room and board, and the waiting lists run into the hundreds of thousands.
Nearly every senior on Medicaid is exempt from the work requirements, so the headline risk isn't what people assume. The real exposure is paperwork, and the people standing around the seniors. A lot of direct care workers earn little enough to be on Medicaid themselves, and long-term care already can't staff itself. That's how a rule aimed at one thing lands squarely on another. The Miller bill is the one I'd watch, because the gap between needing help and affording help is exactly where most families I meet are stuck.
Wonder
10. NASA launched a telescope that sees a hundred times more sky than Hubble
On August 30, the Nancy Grace Roman Space Telescope lifted off at 7:26 a.m. Eastern on a SpaceX Falcon Heavy out of Kennedy Space Center, and separated from the rocket 31 minutes later on its way to L2, about a million miles out. Its primary mirror is the same size as Hubble's, 7.9 feet across, but its 300-megapixel camera takes in roughly a hundred times more sky per shot. NASA says it'll gather data a thousand times faster than Hubble, sending back 2,500 terabytes over five years against Hubble's 172 terabytes in thirty. The mission ran $4.3 billion, and it's expected to turn up more than 100,000 new exoplanets while going after dark matter, dark energy, and the Hubble Tension, which is the unresolved disagreement about how fast the universe is actually expanding.
I always try to end these on something that has nothing to do with money, and this one earns the spot. Same mirror as Hubble, a hundred times the sky, because we got better at the camera. Progress isn't always a bigger machine. Sometimes it's the same machine and thirty more years of learning how to use it. A hundred thousand planets is a number I can't really feel, so I've settled on this instead. For almost all of human history the sky was a fixed decoration overhead. Within my own lifetime we started counting the worlds in it.
That's the ten. Seven weeks of catching up in one sitting, and I'll try not to let it stretch that long again.
If only one of these is worth your time, make it the pancreatic one. It's the story I'd want somebody to tell me about if it were my family.
