Something finally moved in Cupertino. Not a new iPad, not a surprise Mac, not even a “one more thing” press release for a dongle. It is people. And in 2025, the people in charge of AI are basically the product.
What happened
On December 1, 2025, Apple announced that John Giannandrea is stepping down as the company’s senior vice president for Machine Learning and AI Strategy, and will stay on as an advisor before retiring in spring 2026. In the same announcement, Apple said Amar Subramanya has joined as vice president of AI and will report to Craig Federighi. Subramanya will lead Apple Foundation Models, ML research, and AI Safety and Evaluation. Apple also said the rest of Giannandrea’s organization will shift to COO Sabih Khan and services boss Eddy Cue.
Why this matters (and why it is not just HR)
If you have been living with Siri for the past decade, you know the vibe: it is fine for timers, it is weirdly fragile for everything else. Apple has been telling us a “more personalized Siri” is coming, then admitting it is taking longer than expected. When a company that controls the entire iPhone experience does a leadership swap this public, it is basically an admission that the current trajectory is not good enough.
The most interesting part is not the retirement. It is the reporting line. AI now sits directly under Federighi, which is Apple-speak for “software engineering is taking the wheel.” That is a real structural change, not just a new nameplate.
Who is Amar Subramanya?
Apple is not hiring a random manager here. Subramanya most recently ran AI as a corporate vice president at Microsoft, and before that spent 16 years at Google, including leading engineering for Google’s Gemini Assistant. Translation: Apple hired someone who has actually shipped large assistant-style systems at scale, in public, with all the mess that comes with that.
Context: Apple’s AI problem is a trust problem
Apple’s pitch for years has been that it moves slower because it cares more. Privacy, reliability, “it just works,” pick your bumper sticker. Generative AI and modern assistants are basically the opposite: probabilistic, occasionally wrong, and very hard to explain. Apple has to make that fit inside an iPhone-shaped box with expectations it created back when Software Update was a once-a-year event and not a weekly apology.
My take
I like Giannandrea. I also like the idea that Apple can be stubbornly careful. But Siri has been stuck in that uncanny valley where it is always about to get better, and yet never quite does. The Subramanya hire reads like Apple acknowledging that “careful” cannot mean “late forever.”
If Apple wants a next-generation Siri that actually feels modern, it has to do three things: be faster, be more context-aware, and fail less dramatically when you ask something slightly off-script. That is not a marketing problem. That is a systems problem. Putting AI under Federighi and bringing in someone with assistant engineering scars is the most credible signal yet that Apple is treating it like one.
Now comes the hard part: shipping it.