In a fastidiously worded announcement on Monday, Apple mentioned John Giannandrea, who has been the corporate’s AI chief since 2018, is “stepping down” to, properly, not work at Apple anymore. He’ll stick round by way of spring as an adviser.

His substitute is Amar Subramanya, a extremely regarded Microsoft government who spent 16 years at Google, most lately main engineering for the Gemini Assistant. It’s a savvy rent, on condition that Subramanya is aware of the competitors intimately.

The transfer is being characterised as a shake-up. It was seemingly inevitable on reflection. Apple Intelligence, the corporate’s reply to the ChatGPT second, has been stumbling since its October 2024 launch. Opinions have ranged from “underwhelming” to outright alarmed.

Its first months had been a number of the roughest. A notification abstract function meant to condense a number of alerts into digestible snippets generated a sequence of embarrassing, unfaithful headlines in late 2024 and early 2025. Amongst different missteps, the BBC complained twice after Apple Intelligence falsely reported that Luigi Mangione, the person accused of killing UnitedHealthcare CEO Brian Thompson, had shot himself (he hadn’t) and {that a} darts participant, Luke Littler, received a championship earlier than the ultimate even began.

Then there was Siri’s promised overhaul, which turned a black eye for Apple.

A Bloomberg investigation printed in Could revealed the depths of Apple’s AI struggles. As an illustration, when Craig Federighi, Apple’s software program chief, examined the brand new Siri on his personal telephone simply weeks earlier than its deliberate launch in April, he was dismayed to search out that most of the options the corporate had been touting didn’t work. The launch was delayed indefinitely, triggering class-action lawsuits from iPhone 16 consumers who’d been promised an AI-powered assistant.

By that time, Giannandrea had already been sidelined, in accordance with Bloomberg. The information group reported that Tim Cook dinner had stripped Siri from Giannandrea’s oversight totally again in March, handing it to Imaginative and prescient Professional creator Mike Rockwell. Apple eliminated its secretive robotics division from Giannandrea’s management, too.

Bloomberg’s investigation painted an image of organizational dysfunction, with weak communication between AI and advertising and marketing groups, price range misalignments, and a management disaster extreme sufficient that some staff had taken to mockingly calling Giannandrea’s group “AI/MLess.” The report additionally documented an exodus of AI researchers to opponents, together with OpenAI, Google, and Meta.

Apple is reportedly now leaning on Google’s Gemini to energy the subsequent model of Siri, an astonishing and, presumably, humbling twist contemplating the extraordinary rivalry between the 2 corporations that dates again greater than 15 years, throughout cell working programs, app shops, browsers, maps, cloud companies, good dwelling gadgets, and now AI.

Giannandrea got here to Apple from Google, the place he ran Machine Intelligence and Search. At Apple, he oversaw the AI technique, machine studying infrastructure, and Siri improvement.

Now Subramanya inherits these duties, reporting to Federighi with a transparent mandate to assist Apple catch up in AI.

It’s an fascinating second for the corporate. Whereas opponents have been pouring billions of {dollars} into huge AI information facilities, Apple has centered on processing AI duties instantly on customers’ gadgets utilizing its customized Apple Silicon chips, a privacy-first method that avoids accumulating consumer information. (When extra complicated requests require cloud processing, Apple routes them by way of Personal Cloud Compute, servers that promise to course of information briefly and delete it instantly.)

Whether or not that philosophy pays off or whether or not it has completely left Apple behind is an impressive query. Apple’s method comes with clear trade-offs. Amongst them, on-device fashions are smaller and fewer succesful than the huge fashions operating in opponents’ information facilities, and Apple’s reluctance to gather consumer information has left its researchers coaching fashions on licensed and artificial information moderately than the large troves of real-world info that gasoline its rivals’ programs.