Apple announced on December 1, 2025, that John Giannandrea is stepping down as SVP of Machine Learning and AI Strategy and will serve as an advisor before retiring in spring 2026. In the same announcement, Apple named Amar Subramanya vice president of AI, reporting to software chief Craig Federighi.
This is a leadership move with product implications. Apple is working toward a more advanced Siri update expected in spring 2026, and it is consolidating foundation models, machine learning research, and AI safety and evaluation under a single executive sponsor connected directly to the software release organization.
What changed and why it matters
Apple set a clear operating model in its press release.
• Subramanya leads Apple Foundation Models, machine learning research, and AI safety and evaluation.
• The remaining parts of Giannandrea’s organization shift to Sabih Khan and Eddy Cue, signaling a broader realignment of ownership.
• Federighi’s oversight expands, tying AI execution to ship readiness and release accountability.
External reporting connects the change to Siri delivery pressure and timeline slippage, with Apple aiming for a major upgrade in 2026.
The higher-order takeaway is governance
From an outside perspective, consumer AI seems to focus on model capability. Within a company the size of Apple, capability turns into a pipeline challenge. Organizations that deliver reliable assistants treat AI like production infrastructure.
That production infrastructure typically includes:
• A single control system spanning data readiness, training signals, evaluation, and rollout policy
• Integrated gates that decide progression from research readiness to product readiness
• Telemetry that measures live quality, latency, and failure modes across device classes
• Clear decision rights across model, platform, privacy, and UX teams
Apple’s structural changes point in that direction by tightening reporting lines and placing model work and evaluation under a VP role that sits close to the software ship organization.
Why Subramanya is an execution oriented hire
Apple highlighted Subramanya’s background across Microsoft and Google, including leading engineering for the Gemini Assistant at Google, and framed his strength as integrating research into products.
That profile maps to the core risk in assistant programs: the gap between promising internal demos and repeatable end user outcomes.
Implications for CEOs and CHROs running assistant and agent programs
If your roadmap includes an agentic layer or a front door assistant, the hiring lesson is structural.
The role you need owns the entire pipeline:
• Model readiness aligned to product requirements
• Evaluation gates that translate quality into release decisions
• Cross-team handoffs that preserve accountability
• Operations that keep the system improving after launch
A strong leader here acts like a GM of the entire system, with authority over model, platform, evaluation, and release processes.
A practical diagnostic for your org
Where does execution strain surface today in your AI work?
• Evaluation gates
• Decision rights
• Cross-team handoffs
• Telemetry and post-release operations
If those pieces feel fragmented, the fastest path to speed and reliability usually starts with ownership design.