After a wild 3 weeks, OpenClaw joins OpenAI

For the past month, the hottest story in AI was the come-from-nowhere viral breakout of an AI-powered personal assistant originally named Clawdbot, now OpenClaw (after Anthropic sent a cease-and-desist letter on the name.)

OpenClaw was the hobby project of a single man, Peter Steinberger, a genius software developer on sabbatical after cashing out for a rumored $100 million from a company he founded. Burnt out after a punishing 13 years of 24/7 dedication to building his company, Steinberger tried to build a life. Mostly he failed, feeling directionless and empty. Then the release of ChatGPT stirred some embers of interest, and he began wondering why Siri and Alexa were still so stupid and useless in this AI era. Finally, one weekend last November, he decided to hack together a more powerful and useful personal assistant for himself, and put the code out on GitHub, the town square of the open source software movement.

By thinking entirely outside the usual norms of software development, Steinberger created a personal assistant for himself that actually worked. He gave Clawdbot complete access to his computer and his entire online life, including email and messaging apps. He used the bot to check in for flights, reserve tables at restaurants - in effect, to automate the boring parts of his life.

Initially, this hobby project got almost no notice. Then on January 20, a well-known software reviewer gave Clawdbot a glowing review online, and interest began to snowball. On January 26, Clawdbot garnered a GitHub star (equivalent to a “like”) from over 25,000 individuals in that single day. The very next day, Anthropic sent Steinberger a cease-and-desist letter, since “Clawdbot” was an obvious pun on Claude, Anthropic’s chatbot. The project was renamed OpenClaw.

Users began posting the remarkable things that they were able to get the bot to do. One user had OpenClaw negotiate a $4,200 discount on a car purchase via email, while the user was asleep. Another user reported that their OpenClaw bot filed an appeal to an insurance denial without ever being told to do so, but just noticing in the user’s email that this was a task that needed to be done. The project’s GitHub stars rocketed to nearly 200,000, the fastest-growing project in GitHub history.

Steinberger began receiving frantic calls from VCs anxious to back him to build another billion-dollar company around the viral success of OpenClaw. But he had no interest in going back into the meatgrinder of running a startup. On February 14, the feverish rush reached a dénouement - Steinberger announced that he had made an agreement with OpenAI to join the company to guide their efforts to make an intelligent personal assistant. OpenClaw would stay open, supported by a foundation that OpenAI would fund.

Pundits have long predicted that AI would enable a single person to build a billion-dollar company with no other employees. Steinberger may have just done that - in a mere 3 months. OpenClaw has demonstrated that there is enormous interest in AI personal assistants. If Big Tech falters in delivering this product, OpenClaw may become the Linux of the AI era - an open source product that outcompetes all the proprietary solutions.

A summary of Clawdbot’s meteoric rise

Peter Steinberger tells the odd, haphazard way he stumbled into AI fame

GPT 5.2 discovers a new result in theoretical physics

OpenAI’s flagship model, known as GPT 5.2, has recently been credited with discovering a new result in theoretical physics, and has been proposed for co-authorship on the resulting scientific article. A small group of top theoretical physicists were working on a particular property of gluons, the particles that hold atomic nuclei together. Their hand calculations were becoming unwieldy, so they used GPT 5.2 to try to simplify some of the math. GPT ground away at the problem for 12 hours straight, came up with a formula that proposed a new theory, then mathematically proved the formula correct.

GPT 5.2 is not sitting on its laurels. It has already extended its theory from gluons to gravitons (the particles that make gravity work), and it’s working on even further extensions. Last year, humans got a Nobel Prize for inventing AI. The day may come when an AI might win a Nobel Prize on its own.

GPT 5.2 discovers a new result in theoretical physics

The scientific paper: “Single Minus Gluon Tree Amplitudes Are Nonzero”

Spotify developers have stopped coding

A seismic shift is occurring in software development - more and more, AI is writing the code, while the humans set them tasks and manage them. The process is farthest along in the big AI model makers. Both OpenAI and Anthropic claim that their latest models essentially wrote themselves. Now the process is spreading to other tech-heavy companies, like audio streaming giant Spotify. At their most recent quarterly earnings call, Spotify CEO Gustav Soderstrom claimed that their top developers “have not written a single line of code since December”, due to AI.

This sounds dire for software developer jobs, except - Spotify is still hiring software engineers. So is OpenAI, and Anthropic. All of them have dozens of open developer positions that they are actively hiring for. How can this be? The work of software engineers didn’t disappear - it moved. Someone has to decide what needs to be built, prompt the codebot, check its work, coordinate with other teams, and talk to customers and other stakeholders. As AI produces more code faster and faster, it’s not impossible that demand for software engineers could go up, not down.

Spotify developers have stopped coding

400 million idle gaming PCs could be AI’s next big resource

Big Tech is spending ungodly gobs of money building giant data centers to run the predicted tsunami of AI applications. Meanwhile, there is a giant cache of idle GPUs perfect for running AI - the graphics cards in the 400 million existing high-end gaming computers around the world, most of which are idle except for a few hours per day, or even just a few hours per week.  

Some innovative startups like Salad.com have begun creating software that can harness these idle computers during their down periods into a distributed network of AI servers, at a fraction of the cost of dedicated servers in a data center. The individual owners of the gaming computers install Salad’s software on their machines, then receive compensation for any use of their machine by Salad’s customers, who pay Salad an hourly fee for the compute. This can be as low as 10% of the hourly fees charged by the big cloud providers like Amazon, Microsoft, and Google.

Total US data center computer power is expected to more than double over the next few years. The existing power of idle gaming computers is over 100 times that amount.

400 million idle gaming PC’s could be AI’s next big resource

Young people in China spurn marriage for fluffy AI pet companions

Growing numbers of young adults in China are delaying marriage and children, or rejecting them altogether. Chinese birth rates are at approximately half the rate that would keep the population stable, and in fact the nation’s population is shrinking by approximately 3 million persons per year. Observers blame soaring living costs and a hard-charging business culture that idolizes 72-hour workweeks as factors that are persuading young people that marriage and family are not worth the cost.

Instead, more and more young Chinese adults are buying fluffy AI-powered pets that act as emotional companions. Made by large manufacturers by Huawei and small startups like Ropet, these pet companions are designed to be emotionally intelligent, offering conversation, empathy, and even keep digital diaries of their interactions with their owners. It’s companionship without the bother of having to consider the needs of another living creature, human or animal.

Young people in China spurn marriage for fluffy pet companions

E-commerce warehouses are implementing a “Lights Out” night shift

As warehouse robots become more ubiquitous and more capable, e-commerce companies are increasing the speed and volume of delivery without increasing costs by running a “lights out” night shift. Robots are allowed to run autonomously all night, picking items for orders, and positioning them to be packed and shipped by humans at the beginning of the day shift.

E-commerce warehouses are implementing a “Lights Out” night shift

Robot dogs will guard fans at World Cup in Mexico

Security at sporting events is getting a robot upgrade. As a case in point, the 2026 World Cup games in Monterrey, Mexico will employ 4 quadruped robots for additional crowd security. Fitted out with video cameras and communications equipment, the robot dogs are remotely operated by humans, much like drones. They are designed to act as “first responders” to any source of trouble, scouting out the situation so that security forces can respond quickly and appropriately

Robot dogs will guard fans at the World Cup in Mexico

Google’s Isomorphic Labs is inventing new ways to discover new drugs

Google spinout Isomorphic Labs is focused on inventing new ways to discover molecules that can be used to treat human disease. The lab is headed by Demis Hassabis, who also heads Google’s DeepMind AI research division. Hassabis won a Nobel Prize for helping invent AlphaFold, an AI system that predicts the shape, and therefore the functionality, of proteins just from the sequence of the amino acids they are made of. Now Hassabis is moving beyond AlphaFold to a new AI system that is being called a Drug Design Engine (DDE), with 2-3 times the power and accuracy of its predecessor. Hassabis has famously predicted that all human diseases could be cured by AI within 10 years. He’s doing his best to stick to that timeline.

Google’s Isomorphic Labs is inventing new ways to discover new drugs

Hinge Health projects $732 million in 2026 revenue, boosted by AI

Hinge Health, which calls itself an “AI-first” digital clinic for musculoskeletal care, announced its 2025 revenues grew by 50% year over year, while Free Cash Flow tripled, due to AI efficiencies that kept care team costs flat. Hinge Health markets itself to self-insured employers, promising to reduce costs of musculoskeletal disease care by reducing surgeries. The employee with musculoskeletal complaints gets access to an app that uses AI to help gather information about their specific complaint, then loops in a remote physical therapist to approve the diagnosis and the proposed care plan. The app demonstrates the recommended exercises with videos, and can assess the patient’s performance and progress via the camera on their phone. The app also allows contact with the patient’s dedicated care team, and even virtual visits with a professional physical therapist if needed. Hinge projects continued robust growth in 2026.

Hinge Health projects $732 million in 2026 revenue, boosted by AI