Apple has just appointed a new AI chief with expertise from Google and Microsoft following the resignation of John Gianandrea.

In a carefully worded announcement, Apple said Monday that John Gianandrea, who has served as the company’s head of AI since 2018, is “resigning” to no longer work at Apple. He plans to remain as an advisor until the spring.

He will be replaced by Amar Subramanya, a highly regarded Microsoft executive who spent 16 years at Google and most recently led the engineering of Gemini Assistant. It’s a smart hire given Subramanya’s familiarity with the competition.

This movement has been characterized as a revolution. Looking back, it seems like it was inevitable. Apple Intelligence, the company’s answer to the ChatGPT moment, has stumbled since its launch in October 2024. Reviews range from “underwhelming” to downright alarming.

The first few months were the toughest. The notification summary feature, which aims to summarize multiple alerts into easy-to-understand snippets, published a series of embarrassing and false headlines in late 2024 and early 2025. Among other failings, the BBC twice complained after Luigi Mangione, the suspect in the murder of United Healthcare CEO Brian Thompson, shot himself (he didn’t shoot) and after Apple Intelligence incorrectly reported that darts player Luke Littler had won the championship before the final. has begun.

Additionally, there was the promised overhaul of Siri, which was a real pain in the ass for Apple.

A Bloomberg study published in May revealed the depth of Apple’s AI struggles. For example, when Apple’s head of software Craig Federighi tested the new Siri on his phone a few weeks before its scheduled April launch, he was disappointed to find that many of the features the company had touted didn’t work. The release was postponed indefinitely, sparking a class action lawsuit by iPhone 16 buyers who were promised an AI-powered assistant.

According to Bloomberg, Gianandrea was already on the sidelines. The news outlet reported that Tim Cook completely stripped Siri of Gianandrea’s oversight in March, handing it over to Vision Pro creator Mike Rockwell. Apple also removed its secretive robotics division from Gianandrea’s control.

The Bloomberg investigation uncovered a broader picture of organizational dysfunction, including poor communication between AI and marketing teams, budget mismatches, and a leadership crisis so severe that some employees derisively dubbed Gianandrea’s group “AI/MLess.” The report also documents the exodus of AI researchers to competitors such as OpenAI, Google, and Meta.

Apple is reportedly currently relying on Google’s Gemini to power the next version of Siri, a surprising and perhaps humbling development considering the two companies’ fierce competition dating back more than 15 years, spanning mobile operating systems, app stores, browsers, maps, cloud services, smart home devices, and now AI.

Giannandrea joins Apple from Google, where he was responsible for machine intelligence and search. At Apple, he oversaw AI strategy, machine learning infrastructure, and Siri development.

Subramanya has now taken over that responsibility, reporting directly to Federighi with the clear mission of helping Apple catch up in the AI ​​space.

It’s an interesting moment for the company. While competitors are pouring billions into massive AI data centers, Apple is focused on processing AI tasks directly on users’ devices using custom Apple silicon chips, a privacy-first approach that avoids collecting user data. (If a more complex request requires cloud processing, Apple routes the request through Private Cloud Compute, a server that processes data temporarily and promises to immediately delete it.)

The open question is whether that philosophy will work or leave Apple forever behind. Apple’s approach comes with obvious tradeoffs. Among them, on-device models are smaller and less capable than the larger models running in competitors’ data centers, and Apple’s reluctance to collect user data forces researchers to train their models on licensed synthetic data rather than the vast amounts of real-world information that fuels competitors’ systems.