The Model Picker

Apple previewed the Extensions framework at WWDC on June 8. iOS 27 will let any user pick ChatGPT, Gemini, Claude, or Grok as the model running behind Siri, Writing Tools, and Image Playground, through a setting in Apple's own iOS preferences. The picker arrives with the consumer release this fall.

The detail that matters is who chooses. Apple Intelligence's model selection now lives in the user's Settings app, not in any contract anyone else signed. A teenager, a designer, a finance analyst, and a security lead can each route their Siri queries to a different vendor on the same model of phone. The choice is system-wide, set per device, changeable any time, and explicitly open to any AI chatbot the App Store approves once the framework ships.

What Apple shipped isn't a model picker. It's a recognition that vendor selection at the OS layer can't be made by anyone other than the person typing the prompt. There are too many use cases per device, too many model strengths per use case, and too much velocity at the model layer to lock the choice to a contract. Apple just routed the decision to the only party with enough context to make it.

Imagine a college student opening Settings on iOS 27. They see four AI brands they could route Siri to and pick the one their favorite YouTuber recommended last week. The picked vendor gets every Siri query, Writing Tools rewrite, and Image Playground generation that runs through their phone, for as long as they don't go back into Settings and change it. Multiply that by hundreds of millions of users making the same kind of choice, and the AI labs have a new audience whose default they have to win.

The asymmetry is the point. Until now, AI vendor selection at the OS layer was Apple's call, made once for two billion devices through a contract with OpenAI. The Extensions framework moves that choice to the device, and the device is in someone's hand. The vendor Apple locked in by contract still gets paid for the integration work. The vendor the user actually picks gets the actual usage. Those two vendors don't have to be the same firm, and at this scale, the gap between them is the part that matters.

The steelman is that Extensions only affects Apple Intelligence surfaces, which are still a small slice of total AI usage. True. Most AI use today happens through web ChatGPT, Claude, Gemini, Cursor, Copilot, and direct API calls, not through Siri. The Extensions choice doesn't redirect a software engineer's Cursor session or anyone's open ChatGPT browser tab. But Apple shipped this framework at the OS layer, on the device people carry by default, with explicit invitation for the App Store to fill it. Once the pattern works for Siri queries, the next iOS release adds Mail compose, Calendar parsing, Notes search, and Photos editing. The OS-layer AI surface is small today, and it's the surface that grows fastest in 2027.

The thing worth seeing is that the company running the tightest end-to-end product stack in consumer tech just handed AI vendor selection to whoever holds the phone. The vendor decision moves to the surface where the work actually happens. The AI question for the rest of 2026 isn't which model wins in benchmarks; it's which model you keep coming back to once the choice is yours, and what that says about the brand that won your default.

What to Do With This

When iOS 27 lands this fall, the Settings change is yours. Pick the model you actually want, install its app from the App Store once Extensions opens, and switch it on. Use the model long enough on real tasks to know whether you'd pick it again, then check in a month later to see if you still would.

If your phone is company-issued, check whether your IT team's mobile device management policy lets you keep that choice before fall. Most MDM profiles haven't been updated for Extensions yet, which means the default is open. If you own AI policy on your team, the policy update needs to land before iOS 27 reaches your fleet.

Also on the Radar

OpenAI Confidentially Files for IPO

OpenAI confidentially filed a draft S-1 on June 8, eight days after Anthropic filed on June 1. Both top frontier AI labs are now pre-IPO simultaneously, for the first time. For any enterprise mid-renewal, the procurement window just constrained: pre-IPO vendors don't aggressively renegotiate, don't ship roadmap features that contradict their S-1 disclosures, and don't accept terms their underwriters would call material. The leverage window that was open last quarter has closed.

UK Announces Sovereign AI Chip Purchases From British Firms

Technology Secretary Liz Kendall confirmed on June 7 that the UK government will make strategic purchases of AI chips from British semiconductor firms, including £150 million for inference chips this summer, as part of a £1.1 billion plan targeting 5% of the global chip market. For any team shipping AI into UK customer pipelines, where compute physically sits is now a regulated procurement variable, not a backend concern.

Want More Than This Newsletter?

Alcreon publishes a daily AI briefing, long-form dossiers, and an analysis feed for the teams actually shipping AI in production. This newsletter is one read out of the full library.

Read the daily feed or browse the editorials.