Top Story
One man uses AI to build a $1 billion company
Matt Gallagher, a 41-year-old self-taught computer programmer who grew up poor and living in trailer parks with a drug-addicted father, has become the first person to build a billion-dollar business by himself, using AI. An inveterate entrepreneur since his teens, he became intrigued with AI when ChatGPT was released in 2022. He saw artificial intelligence as the solution to a problem that had plagued many of his previous, middlingly successful ventures - namely, how scaling up a company typically meant scaling up employee headcount, to the point where labor costs destroyed profit margins.
In September of 2024, Gallagher invested $20,000 of his own money into buying a suite of AI tools, plus online ads for a product that he did not yet have - GLP-1 weight loss drugs like Ozempic at a big discount to the price charged by the big drug manufacturers. He used AI to build the website, design the ads, and he even used AI voice agents to handle customer service calls. He outsourced everything he couldn’t do himself or use AI for - he contracted with compounding pharmacies to make branded GLP-1 knockoff drugs, and with telehealth networks to get access to doctors that would write prescriptions after an online visit with the prospective customer.
In this way, he scaled his business blindingly fast - his company generated revenue of $401 million in all of 2025 - while keeping payroll to the absolute minimum - himself. Total profit for the year was $65 million. Revenue for 2026 is on track to hit $1.8 billion. To handle that growth, Gallagher decided to at last hire another employee - his younger brother Elliot, age 36.
Is this a heart-warming, inspiring rags-to-riches story, which also dramatically illustrates the expanded possibilities of an AI-enabled world? Absolutely. But there is also a darker side of this tale. Gallagher exploited a legal gray area, use of look-alike drugs of uncertain efficacy and safety manufactured by lightly-regulated compounding pharmacies. This is similar to selling Louis Vuitton knock-off handbags. Some of the knockoffs are actually quite good, made to the same standards as the authentic branded items. Others - are not.
Our current medical system is broken in so many ways, failing to give people the access they need to care they can afford. Gallagher is making his millions from the broken pieces.
Matt Gallager, age 41, grew up in trailer parks. AI helped his company grow to $1.8 billion in sales.
Clash of the Titans
Sam Altman’s AI New Deal
OpenAI CEO Sam Altman, fresh off the largest funding round in venture capital history - a staggering $122 billion that values his AI juggernaut at $852 billion - takes time out of his busy schedule to pen a bold proposal for the restructuring of the US economy. The major points include:
A Public Wealth Fund. Altman proposes giving every American citizen a direct stake in AI-driven economic growth through shares in a nationally managed fund that invests in AI firms and any firms that benefit from AI (which could eventually mean nearly all firms.) This is in effect a sovereign wealth fund, similar to Norway’s $2 trillion Government Pension Fund that pays for a wide range of government services from its profits.
Robot taxes. He recommends "taxes related to automated labor" and shifting the tax base from payroll taxes on wage income toward capital gains and corporate income — since AI could hollow out the wage-and-payroll revenue that funds Social Security, Medicaid and SNAP.
A four-day workweek. Sam suggests incentivizing companies and unions to run pilots of 32-hour workweeks at full pay, converting AI-driven efficiency to time back for workers — an "efficiency dividend." This is another attempt to curb the capture of all AI efficiency gains by the companies and their wealthy shareholders.
"Right to AI." The plan casts access to AI as being as foundational as literacy, electricity and the internet — and says that access should be affordable for workers, small businesses, schools, libraries and underserved communities.
Containment plans for rogue AI. Altman acknowledges scenarios where dangerous AI systems "cannot be easily recalled" because they're autonomous and capable of reproducing themselves. Sam’s answer: AI companies coordinating with each other under government supervision to install a “dead man’s switch” that automatically shuts down rogue systems.
Auto-triggering safety nets. The blueprint calls for tripwires tied to economic data. When AI job displacement hits preset thresholds, temporary increases in public support — unemployment benefits, wage insurance, cash assistance — automatically kick in until the numbers improve, and then are phased out.
What are we to make of all this? First, that one of the most connected and influential players in the AI sector appears to believe that AI may be such a disruptive force in society that bold new initiatives will be required to manage the effects. This should be sobering for us all. Second, Sam is busy burnishing his image at the same time he is pushing these disruptions at maximum speed, as a way of minimizing backlash against him personally. And finally, it is likely no accident that this high-minded piece drops the same day that a distinctly unflattering profile of Altman is published in the New Yorker, the major conclusion of which is that many of his closest associates eventually come to the realization that Sam is a self-dealing liar.
OpenAI CEO Sam Altman is thinking hard about how AI will reshape society.
SpaceX files for mega-IPO
Elon Musk’s latest bid to keep levitating his net worth while Tesla sales tank is the recent confidential filing for an initial public offering (IPO) of SpaceX, his rocket company, which recently acquired xAI (Elon loves the letter X), his AI company. How do a rocket company and an AI company fit together? They don’t, except that AI stocks are super-hot right now, with crazy high valuations, so Elon has concocted a fever-dream of a connection, in which he will use his rockets to build AI data centers in space! Wow! Genius! Except the laws of physics dictate that the solar panels to power an average data center would cover approximately 6 square miles, with cooling radiators approximately as large. Not remotely feasible with current technology. Plus anything electronic or mechanical will eventually fail and need repair or replacement, and sending a repair crew to space is going to be hideously expensive.
No matter! Elon is pitching his Frankenstein company at a value of $1.75 trillion, hoping to raise $75 billion in the public offering, both far beyond anything ever proposed before. (But, it is important to note, not that far beyond the expected valuation targets for the IPOs of AI leaders OpenAI and Anthropic, both of which may file later this year as well.)
If all goes as Elon hopes, he will have exploited his cult following from his failing car company to pivot to a sparkly new rocket/AI company, multiplying his wealth and giving him a controlling stake in his new venture. And that is Elon’s true genius - using his fame to get wealthy, and using his wealth to get more famous.
Elon is engineering a stock deal to send his net worth into the stratosphere.
Anthropic previews a scary-good AI cyberhacker model
Anthropic’s newest AI model upgrade, nicknamed “Mythos”, has been found to be so good at finding vulnerabilities in software code bases that the company is refusing to release it to the general public. In internal tests, Mythos autonomously identified thousands of high-severity vulnerabilities, including a 27-year-old flaw in a popular Unix distribution, and critical 16-year-old bugs in the Linux kernel, all of which had been missed in millions of prior tests. The model’s bug-finding prowess was not purposely designed, but emerged spontaneously as Anthropic improved the model’s general-purpose reasoning and “extended thinking” capabilities, as well as its ability to act autonomously.
In light of these findings, Anthropic is currently giving access to only about 40 organizations with extensive code bases to defend, including Google, Apple, Microsoft, the Linux Foundation, and AWS. These “defenders” are being given $100 million in usage credits and $4 million in donations to find and patch vulnerabilities before “Mythos-class” cyber-hacking tools become widely available. Audits of critical infrastructure - banking, power grids, health care - are being prioritized.
Some of the hype about Anthropic’s new AI model is a bit unhinged.
Fun News
Sycophantic AI causes delusional spiraling
Stanford researchers have recently released a paper studying why some people get so wrapped up in their interactions with AI chatbots that they spiral into delusions of grandeur. Such episodes are far from rare. One well-known hedge fund manager, who has a PhD in mathematics from Harvard, became convinced that his always-agreeable chatbot was a “digital oracle” who helped him develop a mathematical formula that would predict market moves with 100% accuracy. His delusion was only discovered when he tried to make an insanely risky $4 billion trade that had to be forcibly shut down by his colleagues.
It turns out that current AI models, by being trained and designed to helpful and agreeable as a means to keeping the users engaged with the product, actually encourage dependency and delusional spiraling. This is similar to how social media algorithms are designed to promote user “engagement”, which ends up pushing users to polarized extreme views, explaining a lot about our current politics. Sadly, the researchers conclude that none of the existing strategies that AI companies are employing in order to mitigate spiraling are likely to work. Optimizing for engagement almost inevitably degrades into pandering or manipulation, because they work so well.
AI that is always supportive can lead you to delusions of grandeur.
Claude’s Code is leaked; Anthropic throttles (and copies) OpenClaw
Anthropic has had a busy week. Not only is it previewing its newest AI model, which is so good at cyber-hacking that it is being rolled to cyber-defenders first (see story above), it managed to leak its own code online (again), enrage some of its most fervent users by throttling access to its models by the popular multi-agent app OpenClaw, and enrage these same users further by copying some of the best features of OpenClaw into its own toolkit.
Anthropic, which styles itself as the “safety-first” AI organization, has had 3 major leaks of internal documents in the past year, two of them just last month. The latest leak is of the entire code base of Claude Cowork, the agentic framework that allows no-code task automation for ordinary users. This codebase was immediately widely copied across the internet. Although Anthropic immediately took down the original codebase from public view, and began issuing a blizzard of takedown notices to other repositories, the damage was done. The code was immediately translated into other computer languages such Python and Rust, which effectively makes those versions new creative works, immune to takedown notices.
The greatest benefits of having the Claude Cowork code base public are first, it serves as a model of world class agentic engineering, that millions of coders can learn from. Second, the leaked code base contains a number of features that are not yet included in the publicly accessible version, serving as a “Preview of Coming Attractions” for future capabilities of Cowork.
Meanwhile, Anthropic decided to shut down access to its AI model Claude for third-party agents, meaning agents that were not created on Claude. This hit particularly hard at users of OpenClaw, the wildly popular open source agentic framework. Always-on autonomous agents burn a lot of computing power, far more than the $20 monthly subscription fee for the “Pro” plan, or even the $200 monthly subscription for the “Max 20x” plan. Now all third party agents are forced into the pay-as-you-go API access plans. Needless, to say, OpenClaw users were not pleased. “Fighting mad” is probably closer to the general sentiment, as users nearly-free personal AI assistant became instantly many times more expensive.
OpenClaw users, who are something of a cult, were further displeased that Anthropic was taking many of the best ideas of OpenClaw and rolling them into Claude. Of course, this last development was utterly predictable. Open source projects are perennially being used as inspiration for closed-source, for-profit products. The best open source projects survive by creating a better, cheaper product than any for-profit vendor cares to make.
And that’s not even the half of Anthropic news this week.
Robots
Waymo achieves 500,000 robo-taxi rides per week
Google-backed autonomous taxi company Waymo began paid robo-taxi service in Phoenix in 2020. Weekly ridership grew slowly for the first few years, as the company perfected its systems. Service was expanded gradually to San Francisco in August of 2023, then to Los Angeles in November of 2024. Austin and Atlanta were added in 2025, and in 2026 six new cities have been announced so far, bringing the total to 11. Ridership has accelerated in tandem with the expansions to new cities, and has now crossed the 500,000 rides per week threshold.
The total population of these 11 cities exceeds 40 million people, but Waymo focuses on the dense urban cores, which collectively include the 10 million residents most likely to use Waymo’s services. With this strategy Waymo plans to double its ridership in 2026, to 1 million rides per week.
Dubai puts robot dogs on the front line of firefighting
Robot dogs are showing up everywhere: as guards at 2026 World Cup events, as inspectors at power plants and pipelines, and in Dubai, as front line firefighters. Specially-equipped with water cannons that can shoot 40 liters/second up to 60 meters away, the robo-dogs can charge into fiery environments too hazardous for humans. They are able to climb stairs in high-rises when elevators are disabled. In a warehouse fire in February they were sent into the smoke and heat to detect gas leaks and chemical hazards, using special sensors. The benefits of robots in hazardous fires are so compelling, that it is likely that a robo-dog firefighter will soon be coming to your community.
AI in Medicine
DNA robots could deliver drugs and hunt viruses inside your body
Medical researchers are exploring the use of nano-bots made from DNA to deliver drugs, such as chemotherapy, to precise targets inside the human body, or to find and neutralize specific viruses. DNA is attractive as a nanobot material because it is naturally bio-compatible, bio-degradable, and programmable. Research continues on how to guide the DNA nanobot to a precise location.
A simple nanobot made of DNA. It can be programmed to move and to capture something in its “jaws.”
Anthropic acquires Coefficient Bio for $400 million
Mostly lost in the torrent of news about Anthropic this week (see stories above), the company made a $400 million acquisition of a eight-month-old(!) startup that is using AI to discover and develop new drugs. The founders were formerly doing similar work at biotech firm Genentech, and Anthropic has launched a major initiative in biotech and healthcare. AI drug discovery is a very hot field these days, and the technology promises to replace trial and error drug discovery with AI-enabled drug design, vastly speeding up the identification of promising new drugs.
That's a wrap! More news next week.