Apple puts Siri back in play through Gemini, but the proof is still waitlisted

Apple announced Siri AI and new Apple Intelligence features at WWDC 2026, while extending Private Cloud Compute to Google Cloud with NVIDIA GPUs for demanding tasks. After last year's Apple Intelligence disappointment, this is less about the keynote and more about whether Siri can finally survive outside the demo.

Siri AI depends on the user's screen and someone else's cloud

Simon Willison reads the announcement cautiously: after WWDC 2024, he believes new Apple AI promises only when he sees them working. This time, Siri AI looks more technically feasible to him because Apple is licensing a custom Gemini-derived model and running it through Private Cloud Compute.

Apple's security blog also says that for demanding tasks, including agentic tool-use and complex reasoning, it has extended PCC to Google Cloud systems using NVIDIA GPUs. That makes the story different from the original Apple Intelligence pitch: less pure on-device magic, more hybrid architecture with a heavy privacy argument.

Availability matters. Willison writes that the iOS 27 Developer Beta can be installed now, but access to the new Siri AI requires making it through a waiting list. For most users, this is not yet an ordinary capability they can verify.

Developers get a way around slow app-by-app integrations

The interesting part is not the Siri brand. Willison points to vision LLMs reading information from the user's screen. If that works reliably, Apple does not need to wait for every existing app to ship a custom Apple Intelligence integration.

For developers, the second signal is Core AI. Apple is showing a library connected to the PyTorch ecosystem that should help move models onto Apple hardware. That is more practical than another assistant promise because it points to tools developers can actually use.

Apple already shipped one convincing trailer instead of a finished product

Last year's lesson is blunt. Apple can produce a convincing keynote, but users need a feature that survives the messy edges of real apps, permissions, languages, notifications and model mistakes. A vision LLM can bypass missing APIs, but it also adds the risk of reading the screen wrong.

PCC on Google Cloud is technically interesting but delicate to explain. Apple has built trust around control of hardware and privacy. Now it has to show why outsourcing part of inference through Google and NVIDIA does not weaken the promise that makes AI on the iPhone acceptable.

The real test starts with people outside the demo room

The near-term signal is simple: credible reports from people who get off the waitlist and try Siri AI in use. Not in a polished prompt, but while moving through apps where context changes every few seconds.

The second thing to watch is whether Core AI attracts model developers or becomes another elegant system layer used by a handful of demo projects. Apple does not need to win the benchmark. It needs to show that its AI works where people already have the phone in their hand.

Lilith's verdict

Apple does not need another round of keynote applause. It needs the first tired commuter on a train to say something messy to Siri and get the right action instead of another apology.