Apple Just Outsourced Siri to Gemini on June 8. The AIML Org Has 90 Days.

On June 8, 2026, Apple walked on stage at WWDC and did two things in one keynote. It announced that the new "Siri AI" runs on a custom Google Gemini model, and it sent Tim Cook off with a handover toward John Ternus. If you recruit AI talent, those two beats are the same beat: Apple's internal AI organization just got publicly demoted, and the people who built it are now reachable in a way they have not been in a decade.

You have roughly 90 days before Anthropic, Meta Superintelligence Labs, xAI, Google DeepMind, and OpenAI finish the obvious passes. Here's how to think about the pool, the messaging, and the surfaces.

What actually happened on June 8

The "All systems glow" keynote announced a rebuilt Apple Intelligence stack. The headline, confirmed by Apple on stage and corroborated by pre-keynote Bloomberg reporting, is that the new Siri is powered by a custom 1.2-trillion-parameter Gemini model. Apple is reportedly paying Google around $1 billion a year for it, with the model routed through Private Cloud Compute alongside Apple's on-device components.

This is not a side integration. This is Apple's flagship AI surface, the one Craig Federighi has spent two years promising, running on a Google foundation model. Every Apple Foundation Models engineer who shipped on-device work now has to explain, to their manager and to themselves, why the user-facing answer comes from Mountain View.

The same morning, Cook framed his own exit. Bloomberg and the Financial Times have reported Apple is preparing for Cook (who turned 65 in November) to retire in 2026, with Ternus, SVP of hardware engineering, as the leading internal successor. Cook grew the company from roughly $350B to $4T. Ternus is, by every account, a hardware lifer. The signal to AI ICs is unambiguous: the next CEO is a hardware engineer inheriting an AI strategy that is, functionally, "rent it from Google."

The pool is bigger than the named departures

Recruiters keep chasing the same five names. Don't.

Ruoming Pang ran Apple's roughly 100-person Foundation Models team, left for Meta Superintelligence Labs in July 2025 on a package reportedly in the hundreds of millions, and then jumped from Meta to OpenAI in February 2026. Ke Yang, who led the AKI (Answers, Knowledge, Information) team, went to Meta in October 2025. Robby Walker, the longtime Siri exec who ran AKI before Yang, departed earlier in 2025. John Giannandrea is shifting to an advisory role and retiring in spring 2026; Amar Subramanya (ex-Google) now runs Apple Foundation Models, ML research, and AI safety/evaluation under Federighi, while the rest of Giannandrea's old org moves under Sabih Khan and Eddy Cue.

Those are the names every sourcer already has in a spreadsheet. The opportunity is one layer down.

~200

AppleInsider's December 2025 analysis parsed Apple's published machine learning papers and pulled nearly 200 names from only about 4 of 96 pages. The real Apple AIML headcount is in the low thousands, not "a few dozen famous people." Co-author lists on arXiv and machinelearning.apple.com are public. The AIML Residency alumni list is public. The new managers under Zhifeng Chen, Chen himself plus Chong Wang, Zirui Wang, Ching-Cheng Chui, and Guoli Yin, are named in trade press.

This is the IC layer: research scientists, MLEs, evals engineers, training infra, on-device quantization specialists. They are not on the open market yet. They are, however, having a very specific bad week.

Segment Siri ICs separately from Foundation Models researchers

The single most useful framing the rest of the market is missing: there are two demoralized populations inside Apple, and they need different messages.

Foundation Models researchers

These people watched their direct work get bypassed. They've published at NeurIPS, ICML, and ACL under the Apple ML banner. Their argument for staying was always "we'll ship the model behind Siri." That argument died on June 8.

Pitch: technical autonomy, a foundation model program that is not being outsourced, comp that clears Apple's ceiling. Bloomberg's line on the Pang package, that it "far exceeds pay at the company for leaders other than CEO Tim Cook," is now a public, quotable comp ceiling. Any competitor offering meaningful equity is, by definition, above it.

Siri ICs

This is a separate group, and they've been demoted twice in 18 months. In 2025 Apple reorganized Siri under Mike Rockwell, the Vision Pro lead, on the hardware side. In 2026 the brains of the product they own were outsourced to Google. Their org chart says "we build the shell around someone else's model."

Pitch: product surface, real LLM ownership, a roadmap that isn't downstream of a vendor contract. This is a particularly strong group for applied AI teams at Anthropic, xAI, and well-funded startups that need people who have shipped an assistant to a billion devices.

If you're trying to build either of these lists from scratch this week, the bottleneck is not LinkedIn. It is identifying the right ICs by what they've actually worked on. This is exactly the surface Refolk was built for: you describe the person in plain English ("Apple researchers who co-authored on-device LLM quantization papers in 2024 or 2025, currently in the Bay Area"), and you get a ranked shortlist pulled from GitHub, LinkedIn, and the open web, including the machinelearning.apple.com paper graph that most ATS-bolted tools never see.

The 90-day clock

Why 90 days? Three reasons.

First, the Ternus transition. Ternus formally takes over in September. Between now and then, AI ICs are reading the tea leaves and deciding whether to wait for a new strategy or move. Once Ternus announces his own AI org structure in late Q3 or Q4, retention packages will land and the easy yes turns into a hard yes.

Second, the competition is already moving. Bloomberg reported in January 2026 that Apple lost at least four more AI researchers plus a top Siri executive in recent weeks, headed to Meta and Google DeepMind. That was before the Gemini deal was public. The June 8 announcement is jet fuel on a fire that was already burning.

Third, the open-web sourcing surfaces (paper co-authors, residency alumni, GitHub) are easiest to mine right before everyone else mines them. Once the third Information article on "Apple's AI exodus" lands, those names will be in every Gem sequence in the Valley.

The cold email writes itself

The Gemini deal is the opener. You don't need to be subtle.

For Foundation Models ICs: "You shipped on-device foundation models so Siri could answer questions. As of June 8, Siri answers questions with Gemini. We think your work deserves to be the product, not the fallback."

For Siri ICs: "Two reorgs in 18 months. First under Vision Pro, now downstream of a Google contract. We're building an assistant where the model team and the product team are the same team."

For AIML residents and recent grads: "Your residency mentor's org just got restructured under a hardware SVP. We're hiring residents directly into staff-track research roles with published comp bands."

The Gemini deal is not a footnote in your outreach. It is the outreach.
</pull>

Three things to avoid. Don't lead with "are you happy at Apple," because the answer is performative and wastes the reply. Don't pitch "we're like Apple but smaller," because the thing they're leaving is precisely Apple-ness. And don't ignore the residency program; the AIML Residency is a one-year program where grads publish at premier venues and rotate across R&D teams, and the alumni list is one of the cleanest sourcing surfaces in AI.

## Where to actually look

A practical surface map for sourcing Apple AI engineers in the next 90 days:

1. **machinelearning.apple.com.** Every paper has a full author list. Cross-reference against LinkedIn for current title and tenure. This is the canonical Apple AIML directory and Apple publishes it themselves.
2. **The AIML Residency alumni page.** One-year residencies, advanced-degree grads, mentor relationships visible. Many residents convert to full-time; the ones who don't are already in the market.
3. **arXiv co-author graphs.** Apple-affiliated papers from 2023 to 2026. Use the co-author graph to find ICs adjacent to the famous departures. If you sourced Pang's collaborators in July 2025, you have a head start. If you didn't, do it this week.
4. **GitHub.** Apple's MLX project and adjacent open-source contributions surface engineers by handle. Cross-reference commits against paper authorship for high-signal targets.
5. **Conference attendance.** NeurIPS 2025, ICML 2025, ACL 2025, Interspeech 2025. Apple sent a lot of people. Badges and accepted-paper lists are public.

The reason recruiters don't do this is that stitching these five surfaces together by hand takes a week per role. Refolk collapses that into a query: ask for "Apple Siri engineers who shipped speech recognition work between 2022 and 2025 and contributed to MLX," and the cross-surface join happens for you. The second mention is deliberate, because this is the exact workflow where the manual version breaks down.

## What this isn't

This isn't a story about Apple being doomed at AI. Apple still ships the device, the silicon, the OS, and the distribution. The Gemini deal might be a smart commercial decision. Plenty of the ICs will stay because their families like Cupertino and their RSU vests are real.

It is a story about a narrow window where the internal narrative inside one of the world's most prestigious AI orgs got materially worse on a specific date. Recruiting is a game of when, not whether. The "when" is now, and the "who" is the IC layer everyone else is going to mine in October.

Two demotions in one keynote. A hardware engineer about to inherit an AI strategy. A $1B/year Gemini contract that turns Apple Foundation Models into the team that built what the product no longer uses. If you wait for the Bloomberg headline that says "Apple AIML attrition hits X%," you are already late.

## FAQ

### How many people are actually in Apple's AIML organization?

The public number doesn't exist, but AppleInsider's December 2025 pass through Apple's published ML research extracted nearly 200 names from only about 4 of 96 pages of documents. The full addressable population (research scientists, MLEs, evals, infra, applied teams across Siri, Foundation Models, robotics, and Vision Pro AI) runs into the low thousands. The Foundation Models team alone was around 100 people under Pang. Treat this as a sourcing target list of around 1,500 to 2,500 names, not a few dozen.

### Why is the window only 90 days?

John Ternus formally takes over from Tim Cook in September 2026. Between June 8 and September, Apple AI ICs are deciding whether to wait for a new strategy or move. Once Ternus announces his org structure and retention packages land in Q3 or Q4, the easy outbound becomes a hard one. Competitors that move in the June to August window get the conversations; competitors that wait until fall get the counteroffers.

### Should I lead outreach with the Gemini news or the Cook succession?

Gemini, every time. The Cook handover is abstract; the Gemini deal is a direct insult to the work the recipient has personally shipped. For Foundation Models ICs, the line is that their model lost the slot. For Siri ICs, the line is that their product is now a wrapper around a vendor. The succession is useful as a closer ("and the next CEO is a hardware lead"), not as the opener.

### What's the best way to find the non-famous Apple AI engineers?

Start with paper co-authorship. machinelearning.apple.com and arXiv together give you the cleanest list of named Apple AIML contributors, including the IC layer under managers like Zhifeng Chen, Chong Wang, Zirui Wang, Ching-Cheng Chui, and Guoli Yin. Cross-reference against LinkedIn for current title and against GitHub (especially MLX contributions) for engineering depth. If you don't want to do the join by hand, this is exactly the kind of multi-surface query Refolk handles in a single ask.