Apple has appointed a fresh AI leader, drawing on experience from Google and Microsoft, following John Giannandrea’s departure from the role.

In a carefully crafted statement released Monday, Apple announced that John Giannandrea, the company’s AI head since 2018, is “stepping down,” effectively ending his tenure at Apple. He will remain as an advisor through the spring.

Taking his place is Amar Subramanya, a well-respected Microsoft executive with 16 years of experience at Google, where he most recently spearheaded engineering for the Gemini Assistant. This hire is considered strategic, given Subramanya’s deep knowledge of Apple’s competitors.

The move is being described as a significant restructuring. In hindsight, it seemed almost inevitable. Apple Intelligence, the company’s response to the rise of ChatGPT, has struggled since its debut in October 2024. Reviews have been largely negative, ranging from “underwhelming” to deeply concerning.

Its initial months were particularly challenging. A notification summary feature, designed to condense numerous alerts into manageable snippets, generated a string of inaccurate and embarrassing headlines in late 2024 and early 2025. Among other errors, the BBC twice reported complaints after Apple Intelligence incorrectly stated that Luigi Mangione, the accused killer of UnitedHealthcare CEO Brian Thompson, had shot himself (which was false) and that darts player Luke Littler had won a championship before the final match concluded.

Then there was the highly anticipated Siri overhaul, which became a point of embarrassment for Apple.

A Bloomberg investigation published in May highlighted the extent of Apple’s AI challenges. For example, just weeks before the planned April launch, Apple’s software chief, Craig Federighi, tested the new Siri on his own phone and was disappointed to find that many of the advertised features were non-functional. The launch was delayed indefinitely, leading to class-action lawsuits from iPhone 16 purchasers who had been promised an AI-driven assistant.

According to Bloomberg, Giannandrea had already been removed from key responsibilities by that time. The news outlet reported that Tim Cook had taken Siri away from Giannandrea’s oversight in March, assigning it to Vision Pro creator Mike Rockwell. Apple also transferred its secretive robotics division from Giannandrea’s control.

Bloomberg’s investigation portrayed a picture of internal disarray, characterized by poor communication between AI and marketing departments, budget discrepancies, and a leadership vacuum so profound that some employees had started sarcastically referring to Giannandrea’s team as “AI/MLess.” The report also detailed a departure of AI researchers to rival firms, including OpenAI, Google, and Meta.

Apple is now reportedly relying on Google’s Gemini to power the next iteration of Siri, a surprising and likely humbling development given the long-standing and intense competition between the two companies, spanning over 15 years across mobile operating systems, app stores, browsers, maps, cloud services, smart home devices, and now, AI.

Giannandrea joined Apple from Google, where he managed Machine Intelligence and Search. At Apple, he was in charge of the AI strategy, machine learning infrastructure, and the development of Siri.

Now Subramanya assumes those duties, reporting to Federighi with a clear objective to help Apple close the gap in AI.

This is a critical juncture for the company. While its competitors have invested billions in extensive AI data centers, Apple has prioritized processing AI tasks directly on users’ devices using its own Apple Silicon chips, a privacy-focused strategy that avoids collecting user data. (For more complex requests that require cloud processing, Apple uses Private Cloud Compute, servers that are designed to process data temporarily and delete it immediately.)

Whether this approach will prove successful or permanently leave Apple lagging behind remains to be seen. Apple’s method presents clear compromises. Among these, on-device models are smaller and less powerful than the vast models operating in competitors’ data centers, and Apple’s reluctance to gather user data has forced its researchers to train models using licensed and synthetic data rather than the massive quantities of real-world data that power its competitors’ systems.