Back in May, it seemed like OpenAI was close to its largest-ever acquisition: a $3 billion deal to absorb Windsurf, a popular A.I. coding company. But on Friday, Google announced an agreement to hire Windsurf’s co-founder and C.E.O., Varun Mohan, plus other key Windsurf employees, in a deal reportedly worth $2.4 billion. (Google is also getting a nonexclusive license for Windsurf’s technology.) Then, on Monday, A.I.-coding startup Cognition—the makers of the much-hyped Devin product—announced that it was acquiring Windsurf’s remaining assets. Jeff Wang, who was instated as Windsurf’s C.E.O. three days ago, described those 72 hours as “the wildest rollercoaster ride” of his career. No kidding.

Of course, acqui-hires are increasingly popular among the leading giants in the A.I. space: Google did it last year with Character.AI, Microsoft did it last year with Inflection, and Amazon also did it last year with Adept. Still, it’s remarkable how quickly the talent war is upending the venture ecosystem, in which investors and early employees bet on a big exit for the company, not just its founders or star employees.

Two attorneys representing MyPillow C.E.O. Mike Lindell in a recent defamation case were sanctioned and fined last week by a federal judge after they were caught using a generative A.I. system to prepare a court filing, which included citations to cases that don’t exist. The lawyer-gets-busted-using-A.I. trope has been playing out almost from the moment ChatGPT burst onto the scene. One researcher has assembled a database of such scenarios that includes more than 200 cases. Somehow, they never learn. Maybe the fines are too low?

And now for the main event…

The feverish integration of A.I. tools across the medical industry has become a focus for virtually every big tech player in the space. Microsoft is pursuing “medical superintelligence”; OpenAI says that “improving human health will be one of the defining impacts of A.G.I.”; Nvidia is “transforming” healthcare with A.I.; Amazon is “reimagining primary care” with A.I.; and Google is “transforming” healthcare. All are ostensibly focused on personalizing medicine with better analytics and predictive diagnostics; enhancing productivity to overcome staffing shortages; and improving operational efficiencies to address burnout while improving patient care.

The excitement has penetrated Washington, D.C., too. Secretary of Health and Human Services Robert F. Kennedy Jr., who heralded the arrival of a so-called “A.I. revolution” at the department, has declared that “President Trump is determined to end the hemorrhaging of rural hospitals, and he’s asked me to do that through the use of A.I.” It all sounds compelling, but things haven’t exactly gone according to plan on the ground. A 2024 survey from National Nurses United—the country’s largest union of registered nurses—found that only 40 percent of respondents trust their institutions will implement A.I. with patient safety “as their first priority.” The survey details cases of missing patient information, and inaccurate algorithmic analyses that “often contradict and undermine nurses’ own clinical judgment and threaten patient safety.”

Michelle Mahon, the director of nursing practice at NNU, told me that while nurses are not anti-tech, their primary concern is whether a given technology will actually improve patient care. “There is no demonstrated effectiveness” when it comes to A.I., she said. “It is not proven to be safe and effective prior to its implementation, and this has caused disruptions.” When asked about the technology’s propensity to “hallucinate,” she quickly noted that “we are talking about people’s lives. Even a small aberration in facts can be life-threatening.”

As with the electronic health record, Mahon said that A.I. is “primarily designed to capture revenue. There’s been a long-standing goal of the industry to reduce labor costs by cutting back on nursing.” As long as the goal involves saving money by “distilling care,” she said, no technological intervention is “going to work in the best interest of patients.” Citing Kennedy’s statements on the supposed capabilities of “A.I. nurses,” she said she does not see a path to the integration of A.I. technologies that would overcome or even address any of these concerns: “This is a red line for us.”

At the heart of these concerns is the slightly more abstract, somewhat longer-term threat of “cognitive offloading” and skill degradation associated with the use of generative A.I. Indeed, several studies this year have identified a decrease in critical thinking associated with using and trusting the technology. According to Dr. Rahul Goyal, Elsevier’s chief medical officer, an overreliance on A.I. in a medical context could potentially cause healthcare workers to lose their “knack.” While “A.I. is a brilliant servant,” he told me, it’s “not the best master.”

When it comes to the question of skill degradation—in which clinicians might become trained to respond to A.I.-powered alerts rather than their own intuition—the threat is multipronged, and dovetails with the problem of cognitive offloading. To wit: When tools seem to work only some of the time (or if the A.I. experiences a hallucination), clinicians in a crisis might not have the skills necessary to address a problem. And even if the tech worked perfectly all the time, the risk of overreliance doesn’t go away.

Last year, the Ascension health system lost access to its electronic health record for more than a month due to a ransomware attack. Mahon said that downtime systems weren’t in place, and staff didn’t know how to work from a physical chart. “As medical healthcare professionals, we are taught immediately, first and foremost, to treat the patient and not the monitor,” she told me. “What is happening with a person is more important than the data. This flips the script and emphasizes the value of data over all other kinds of knowledge. And that is dangerous.”

There were lots of great responses to my first two columns last week. Here’s a sampling of the feedback in my inbox…

“I do not agree with this premise that videos or video games will eventually be generated for each individual user. People crave collective experiences, something they can engage other human beings about. If everyone is watching their own unique video or playing their own specially tailored game, it removes the ability to engage with others. Can someone who works in A.I. explain to me why they think people won’t care about the collective experience in favor of something uniquely for them? This is a deep misreading of why people engage with things, and I find it frustrating that these experts never comment on that while claiming this is the future of media.” —A film industry person

“Your most recent newsletter includes a Quote of the Week from someone named Alexandra Ebert at Mostly AI. Her remarks reflect the usual ignorance of history of many of our tech innovators. The early days of car safety did not involve a person with a red flag walking around to warn passengers, with seat belts coming online downstream. Once cars were commercialized, it became necessary to construct literal guardrails on roads. And traffic lights. And stop signs. And road infrastructure. As individuals it behooves us to take precautions to drive carefully: watch out for pedestrians and other cars, maintain a safe speed, and so on. But we shouldn’t expect that our personal decisions alone will be enough to ensure road safety.” —A longtime Puck subscriber

See you Thursday,
Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St, New York, NY 10006