Your iPhone Now Has Three AI Options. The Skill That Matters Isn't Model-Picking.
With iOS 27, you can route Siri through Claude, ChatGPT, or Gemini. That choice is the easy part. The skill is knowing how to frame the ask in the quick moments — before a meeting, between tasks, on a walk.
By Forge Team
For most of the last three years, using AI well required some deliberate setup: open the app, write a clear prompt, review the output. That context — sitting down with the intention to do AI work — shaped how the skill was taught. Starting this month, that context is one of many. The more common context is the two minutes you have between your car and the office door.
What Apple announced
At WWDC26 (June 8–9), Apple shipped a ground-up rebuild of Siri powered by Google Gemini. The more consequential announcement was in the developer session details: iOS 27 includes an "Extensions" framework that lets users set Claude, ChatGPT, or Gemini as the AI engine behind any Apple Intelligence feature — voice, text, summarization, on any Apple device. One settings toggle. No third-party app required.
Ben Thompson described the positioning as integrated and reliable versus powerful but unpredictable — a direct contrast with the Fable 5 controversies of the same week. (Stratechery, June 13, 2026.) Developer Simon Willison called it technically credible in a way the 2024 Apple Intelligence rollout never was. (Willison, June 8, 2026.)
AI stopped being an app you open with intention. It is now ambient: in your phone, your car, your AirPods, your Mac. The Extensions framework means you can pick your model — but the more interesting question is what to do with it when you have thirty seconds and no keyboard.
What changes on Monday
The deliberate session workflow — careful prompt, long context, iterative refinement — is still the right approach for analysis, drafting, and anything requiring more than one exchange. That isn't going away.
But a lot of professional AI use is quick: get me a first take on this, help me think through this before I walk in, summarize this in one sentence. Those tasks happen in moments. And in moments, the gap between useful and useless output is almost entirely about how well the ask was framed — not which model was chosen.
"Summarize this email thread" gets a summary. "Summarize this email thread: what's the disagreement, and what decision is being avoided — I'm meeting with the two parties in ten minutes" gets something you can use in the room.
Ambient AI doesn't reduce the framing requirement. It increases it, because the brief is verbal, fast, and typed with one thumb. Vague ambient prompts produce generic ambient answers. The skill is compressing enough context into a quick ask to make the output worth having.
Priya: the marketer who started using AI in the car
Priya runs marketing at a 35-person legal tech company. She had been using Claude for strategic briefs and campaign planning — long, intentional sessions with careful prompts and follow-up questions.
After the iOS 27 update, she started using Siri with Claude extensions in quick moments: in the car between calls, during the ten minutes before a client meeting, on her morning walk. The first week was frustrating. The quick outputs felt thin.
She diagnosed the problem after comparing her desktop prompts with her phone prompts. On her laptop: "Summarize what a compliance officer at a 500-person company would be most skeptical about when evaluating a new legal tech vendor — I'm preparing a case study draft." On her phone: "Summarize this article." Same model. Very different output.
She now treats every quick ask the same way: one sentence for the task, one sentence for the context, one sentence for the outcome she needs. It takes five more seconds to compose. The output is usable instead of generic.
Practice the quick-context frame: what the task is, why it matters right now, and what a useful answer looks like — in under three sentences.
When the model actually matters
The Extensions framework does create a real decision — not for ambient quick tasks, but for the sessions where depth is the differentiator.
David is an operations director at a 60-person manufacturing firm. He uses Gemini through Siri for ambient tasks: quick research, meeting prep, text summaries while walking between the floor and his office. Those tasks reward reliability and speed. He uses Claude on his laptop for the quarterly analysis work — pulling 200 pages of supplier contracts into context, reasoning across them, producing a recommendation he'll stake his name on.
The choice isn't "which AI is best." The choice is "what kind of task is this, and which tool is built for that?" For ambient, quick, low-stakes tasks, the built-in model is usually enough. For high-stakes, high-context work that requires careful reasoning and long-form output, the model choice matters a lot.
The Extensions framework is useful precisely because it forces that distinction. You can't have one ambient setting without thinking about what you're routing there and why.
Map which of your AI uses are quick and ambient versus deep and deliberate — and decide whether your current tool setup matches that distinction.
The thing worth deciding now
AI on your phone means more opportunities to use AI — and more opportunities to use it carelessly. The Extensions setting takes ten seconds to configure. The habit of framing the ask well takes longer to build.
Pick one quick AI use you do regularly: a pre-meeting summary, a first-take on a message, a question you'd normally Google. Try framing it with task, context, and outcome, instead of a single command. See whether the output changes. If it does, that's the skill.
Identify one ambient AI use in your current work and decide whether it needs more deliberate framing — or whether the quick version is already good enough for the stakes.
Like this post?
Get the next one in your inbox. Practical AI skills, no filler.