Apple has initiated a major restructuring of its artificial intelligence division, announcing that longtime executive John Giannandrea will step aside after nearly seven years at the helm. He will remain as an adviser until spring, while Amar Subramanya, a seasoned leader with roots at Microsoft and 16 years at Google, steps into the top role. Subramanya, who most recently oversaw engineering for the Gemini Assistant is seen as a strategically savvy appointment, bringing direct insight into the competitors Apple is now racing against.

The leadership shift comes at a fragile moment for Apple Intelligence, the company’s flagship AI initiative. Since its rollout in October 2024, Apple’s system has endured an onslaught of criticism, with early reviewers describing it as underwhelming, and in some cases, alarming. A botched notification-summarization tool triggered public embarrassment by producing false news blurbs, including a claim that Luigi Mangione—accused in the death of UnitedHealthcare CEO Brian Thompson—had taken his own life, and another incorrectly declaring darts prodigy Luke Littler as a champion before the final match had been played.

Compounding these problems was Apple’s highly anticipated rebuild of Siri, which became an internal crisis. A Bloomberg investigation in May found that many of the assistant’s promised features failed to work when tested by software chief Craig Federighi just weeks before its scheduled April debut. The launch was indefinitely postponed, sparking class-action lawsuits from iPhone 16 buyers who had been promised an upgraded, AI-equipped voice assistant.

The Bloomberg report also revealed deep fractures within Apple’s AI organization. Ahead of the most public failures, Giannandrea had already been marginalized, with Tim Cook moving Siri under Vision Pro architect Mike Rockwell and removing the company’s secret robotics initiative from his oversight. The investigation described a division plagued by poor communication, mismatched budgets, and low morale, so strained that some employees reportedly joked that the AI group had become “AI/MLess.” A steady departure of researchers to rivals including Meta, Google, and OpenAI further weakened the unit.

In a striking reversal given the companies’ fierce history, Apple is now preparing to rely heavily on Google’s Gemini technology to power the next generation of Siri, a move seen as both pragmatic and surprising. Subramanya will now take on the enormous responsibility of steering Apple’s AI strategy, reporting directly to Federighi and tasked with restoring momentum at a time when the company is struggling to keep pace.

This transition arrives as Apple continues to defend its distinctive AI philosophy. Instead of massive cloud data centers like those run by Google and Microsoft, Apple has championed a privacy-first approach that keeps most processing on the device via Apple Silicon. For tasks that exceed on-device limits, the company uses Private Cloud Compute, a system designed to handle requests temporarily and delete data immediately. The model is secure—but also restrictive. On-device AI models are necessarily smaller and less powerful, and Apple’s refusal to harvest extensive user data forces its researchers to rely on licensed or synthetic datasets. Whether this principled path will sustain Apple’s competitiveness—or leave it lagging behind rivals moving at full scale—remains one of the company’s most pressing questions.