This week Ksenia Se, host and founder of Turing Post, argued that AI is moving beyond conversation and into the operational loops where actual work is done. Recent developments — SpaceX’s $60 billion stock purchase of Anysphere (the company behind Cursor), G7 discussions about access to frontier models, and Midjourney’s pivot into medical hardware — all point toward a contest to own the infrastructure tying agents, data and repeated measurement together.
When agents own the loop, the IDE becomes infrastructure
SpaceX’s acquisition of Anysphere, reportedly for $60 billion in stock, looks simple at first glance until you examine what Cursor actually is. Cursor is a popular AI‑assisted code editor, but Ksenia warned that viewing the deal only through that lens is too narrow, especially from Elon Musk’s perspective. SpaceX may be positioning Cursor to become the new center of software work the way GitHub was in the previous era.
Under the old model, GitHub owned the pull request. In the new model, however, ownership of the full loop — agents reading a repo, writing code, opening pull requests, running tests, handling failures and enforcing engineering standards — remains unsettled. GitHub still owns the system of record and is defending its position: Chief Product Officer Mario Rodriguez told Turing Post that GitHub’s mission has shifted from human‑developer collaboration to developer‑and‑agent collaboration, making the platform agent‑native across APIs, UX and underlying infrastructure.
Ksenia countered that “Cursor’s advantage is that it owns the developer’s active coding surface,” where work begins. If agents will write more code than humans, software infrastructure should be designed around agents from the outset. Cursor was built for agents; GitHub was built for humans and is now catching up. That architectural choice may matter more than any single product feature.
Frontier AI access is turning into a geopolitical issue
At this week’s G7 summit delegates discussed a “trusted partners” framework that would allow select allied nations access to advanced US AI models. The talks follow a US order that restricted foreign nationals’ access to Anthropic’s frontier systems on national security grounds. Models that can write software, discover vulnerabilities and operate across tools are capability systems rather than mere productivity applications, and access rules are evolving to reflect that.
For a long time AI regulation focused on questions like how to label synthetic media, reduce hallucinations, prevent bias and make chatbots safer. Now the debate is far broader: who can use these capable systems? Can allies use them? Can cybersecurity firms outside the US use them? Can non‑US employees at US labs have access? Can European companies use American models that may also be strategically sensitive? This has moved beyond traditional software licensing into capability access control.
The underlying tension is the dual‑use problem: a model capable enough to find software vulnerabilities for defense can also be used to find vulnerabilities for offense. The “trusted partners” concept reflects a new AI geopolitics in which states jockey to secure strategic advantages for themselves and their allies. It applies access structures once reserved for physical military hardware to software capabilities that are too strategically important to be fully open and too useful to keep wholly locked down. As Ksenia put it, the alliance is “not literally NATO, but [it is founded on] the same kind of logic.”
Access restrictions may also affect the talent that built these systems, who increasingly are not citizens of the countries seeking to control them. For example, AI researcher Andrej Karpathy, recently hired by Anthropic, is publicly described as Slovak‑Canadian. If access controls extend to non‑US citizens, he and others like him could be denied access to the very systems they are employed to work on.
AI is entering the measurement loop
Midjourney, known for AI‑generated images, announced a new medical division and a full‑body ultrasound scanner based on water immersion, developed in partnership with Butterfly Network. The device is designed to scan the entire body in 60 seconds: a person descends into a shallow pool on a motorized platform and passes through a ring of roughly half a million ultrasound sensors, each acting as both transmitter and receiver. The system uses over two petaflops of processing power to reconstruct a 3D body map from returning wave data. Midjourney says the resulting images look comparable to current MRI output at a fraction of the cost and time, though that claim still requires rigorous clinical validation.
According to a disclosure from Butterfly Network, the current prototype uses 40 Butterfly ultrasound‑on‑chip devices per system; Butterfly confirmed its codevelopment and licensing agreement with Midjourney. Midjourney plans to open a facility in San Francisco in 2027 that embeds the device in a spa environment alongside hot tubs, saunas and cold plunges. Diagnostic medical uses will require FDA approval; the initial focus is on body composition mapping.
If Midjourney can assemble a library of full‑body scans collected over months and years, that longitudinal record would provide doctors and AI health tools with baseline data that today typically exists only in clinical trials. This is the same structural logic that Ksenia traced through Cursor and GitHub: value compounds inside the loop through repeated, precise measurement over time. Midjourney is positioning itself to own that loop in the health domain.
What’s next
The competition for AI advantage is shifting from model capability to infrastructure position. Who owns the coding loop? Who controls access to frontier systems? Who builds the measurement environments where health data accumulates over time? These are questions about where intelligence meets operational reality, not simply which model tops a benchmark.
Hiring moves this week underscore how seriously labs are treating this phase: John Jumper, who shared a Nobel‑related prize with Demis Hassabis for AlphaFold work, left Google DeepMind for Anthropic. Noam Shazeer, a coauthor of “Attention Is All You Need,” reportedly left Google for OpenAI after Google paid roughly $2.7 billion in 2024 to bring him back.
Next week Andreas Welsch will return to discuss multi‑vendor strategy with Conductor’s Matt Palmer and cover a range of industry moves. Starting in July, registration for the live event will be open only to O’Reilly members; those interested can try O’Reilly free. Radar will continue to publish weekly takeaways on Fridays and share full episodes on YouTube, Spotify and Apple.
