Apple announced on Monday that John Giannandrea, its AI chief since 2018, is stepping down from his role while remaining as an adviser through spring. The company appointed Amar Subramanya, a former Microsoft executive with extensive Google experience, as his replacement to address ongoing challenges in its AI initiatives.
John Giannandrea joined Apple in 2018 after serving at Google, where he led Machine Intelligence and Search divisions. At Apple, he directed the overall AI strategy, managed machine-learning infrastructure, and supervised Siri development. His departure marks a transition in leadership for these areas, with Subramanya now assuming those duties.
Amar Subramanya brings significant expertise to the position. He worked at Microsoft as an executive and previously spent 16 years at Google. In his most recent role at Google, he led engineering efforts for the Gemini Assistant. Subramanya reports directly to Craig Federighi, Apple’s software chief, and takes on the responsibility of advancing Apple’s AI capabilities.
Apple Intelligence, Apple’s AI product suite, launched in October 2024 as a response to advancements like ChatGPT. Since its debut, the system has faced various criticisms. Reviews described it as underwhelming, with early performance issues drawing particular attention from users and media outlets.
One notable problem occurred with the notification summary feature, designed to condense multiple alerts into shorter messages. In late 2024 and early 2025, this feature produced inaccurate headlines. For example, it falsely claimed that Luigi Mangione, accused of killing UnitedHealthcare CEO Brian Thompson, had shot himself. The BBC reported this error, highlighting the misinformation generated by the system.
Another incident involved the same feature reporting that darts player Luke Littler had won a championship before the final match concluded. The BBC complained twice about these inaccuracies, underscoring the reliability concerns with Apple Intelligence during its initial rollout.
The promised overhaul of Siri represented a major component of Apple’s AI plans. A Bloomberg investigation published in May 2025 detailed significant setbacks in this project. The report revealed internal challenges that prevented the timely delivery of enhanced features for the virtual assistant.
Craig Federighi tested the updated Siri on his personal iPhone just weeks before the planned April launch. He discovered that several promoted features failed to function as expected. As a result, Apple postponed the release indefinitely, leaving the future timeline uncertain.
This delay prompted legal action from consumers. Class-action lawsuits emerged from iPhone 16 buyers, who had purchased the devices based on Apple’s assurances of an advanced AI-powered Siri. The suits claim that the company failed to deliver on these commitments.
By March 2025, Tim Cook, Apple’s CEO, had restructured oversight of AI projects. He removed Siri entirely from Giannandrea’s responsibilities and assigned it to Mike Rockwell, the creator of the Vision Pro headset. Cook also transferred control of Apple’s secretive robotics division away from Giannandrea’s group.
The Bloomberg investigation exposed broader organizational issues within Apple’s AI efforts. Communication between the AI team and marketing personnel proved weak, leading to misaligned expectations. Budget allocations did not align with project needs, contributing to inefficiencies.
A leadership crisis affected the AI division under Giannandrea. Some employees referred to the group as “AI/MLess” in jest, reflecting perceptions of limited progress. The report noted an exodus of AI researchers, with many departing for positions at competitors including OpenAI, Google, and Meta.
In response to these challenges, Apple has begun integrating external technology. The company plans to use Google’s Gemini model to power the next iteration of Siri. This decision contrasts with the historical competition between Apple and Google, which spans over 15 years across areas such as mobile operating systems, app stores, browsers, maps, cloud services, smart home devices, and now artificial intelligence.
Apple’s AI approach emphasizes processing tasks on users’ devices using custom Apple Silicon chips. This method prioritizes privacy by avoiding the collection of user data. For more complex queries, requests route through Private Cloud Compute, where servers process the information temporarily and delete it immediately afterward.
This on-device strategy results in models that are smaller in scale compared to those operated in competitors’ data centers. Apple’s commitment to not gathering user data limits training resources. Researchers rely on licensed datasets and synthetic data, rather than the extensive real-world information used by rivals to develop their systems.
