Today at WWDC 2026, Apple CEO Tim Cook announced the new Siri: rebuilt from the ground up, finally useful, with its cloud features powered by a custom 1.2-trillion-parameter Google Gemini model under a $1 billion-per-year licensing deal. Apple's own on-device models still handle local tasks — but the heavy lifting is rented.
Let that settle in. Apple — the most vertically integrated company in consumer technology, the company that designs its own chips, writes its own operating systems, and has spent a decade telling investors it owns the full stack — decided that frontier AI intelligence is something you buy from Google.
That's not a product decision. It's a strategic admission. And it has direct implications for every engineering org currently trying to staff an "AI team."
What Apple Actually Announced
The rebuilt Siri features a chat-style interface, Dynamic Island integration, full personal context (emails, photos, documents), and cross-app actions that actually work. iOS 27's new Extensions system lets users select ChatGPT, Gemini, or Anthropic's Claude as their preferred AI backbone.
Former AI head John Giannandrea is out. Mike Rockwell, who led the Vision Pro effort, now leads the Siri rebuild. Apple's VP of AI, Amar Subramanya — previously at Google and Microsoft — oversees model research. And the model doing the cloud work is Google's.
This is Tim Cook's final WWDC as CEO. On September 1 he becomes Executive Chairman and hands John Ternus a company where the core cloud AI layer is licensed, not owned.
The Build-vs-Buy Question Nobody Is Asking Honestly
For three years, tech companies have been hiring "AI teams" with the assumption that proprietary model development was the strategic moat. The pitch to the board: we need ML engineers, we need a model, we need to own this capability.
Apple had more resources to build than any company in enterprise software. It had thousands of ML engineers, billions in research budget, and three and a half years of runway after ChatGPT made this a board-level priority.
The result: a Siri so bad it became a meme. So they bought Gemini.
"The incomplete AI strategy is still the biggest overhang, but we think Apple still has approximately 1.5 years to effect a compelling solution."
— Krish Sankar, analyst, TD Cowen, NBC News/CNBC
If Apple couldn't make the build-it-yourself thesis work at that scale, most companies trying to staff an internal LLM team for their SaaS product or enterprise application are kidding themselves.
The question isn't "should we build our own model?" That question is answered. The question is: what AI capability is actually worth owning — and what should just be a vendor relationship?
What This Means for Your Hiring Roadmap
Here's where I see CTOs getting the talent strategy wrong right now.
They're still trying to hire foundational AI researchers when what they actually need is AI integration engineers — people who know how to take a frontier model API, connect it to internal data systems, wrap it in appropriate governance, and ship it in a way that doesn't blow up your compliance team.
That's a different skill set than training models. It's closer to platform engineering than research (think the person who wired Stripe into your billing stack, not the person who wrote the paper on payments). And it's in shorter supply than anyone is saying publicly, because the job boards still describe it in research terms.
"In numbers, there's probably going to be significantly more AI Engineers than there are ML engineers / LLM engineers. One can be quite successful in this role without ever training anything."
— Andrej Karpathy, via Twitter, quoted in The Rise of the AI Engineer, Latent.Space (June 2023)
Apple's Gemini deal is a public proof point that you don't need to own the intelligence layer. You need to own the integration, the context, the user experience, and the trust framework around it. Apple's value isn't in Gemini's weights — it's in the fact that Siri now knows your calendar, your emails, your files, and can act across every Apple app in a way Google Assistant can't.
The moat is the context and the integration. The model is rented.
The Staffing Implication
If you're building a job description right now for an "AI Engineer" or "Head of AI," I'd push you to answer four questions before you post it:
- Are you actually training models, or are you integrating them? Be honest. Most companies are doing the latter and hiring for the former.
- What does your AI integration stack look like in 18 months — and does the person you're hiring know how to build that, or just advise on it?
- Is the bottleneck research, or is it getting your internal data clean enough for a model to use? (Hint: for most enterprises, it's the data.)
- Who owns AI governance in your org — and does your AI hire know how to work inside that structure, or just around it?
Apple spent years hiring the wrong kind of AI talent for what it actually needed to build. The Giannandrea era produced beautiful research and a broken product. The Rockwell era is licensed.
Don't make Apple's mistake at 1/100th of Apple's budget.
At VC5 Consulting, we help engineering and operations orgs build the right AI talent strategy — which usually starts with being honest about what you're actually building. If your current AI hiring plan doesn't answer those four questions, we should talk.