
如果你一直在关注苹果向“AI优先”公司的缓慢转型,那么App Store核心掌门人菲尔·席勒(Phil Schiller)的离开并不会令人感到意外。多方报道指出,席勒选择退居二线,是因为在从生态系统中榨取更多经常性收入的计划上,与热门继任CEO人选约翰·特纳斯(John Ternus)产生了分歧。换句话说:准备好为iPhone上的每一个AI助手按月支付订阅费吧。
作为每天都在测试AI智能体的人,我眼睁睁看着App Store变成了高价“套壳应用”的坟场。但在特纳斯领导下,苹果的宏伟蓝图远不止于从第三方AI初创公司抽取30%的抽成,而是要在操作系统层面将这种“AI税”制度化。我们已经看到苹果在为高阶版Apple Intelligence奠定基础。想让Siri在不产生幻觉的情况下真正执行复杂的跨应用工作流?那可能每月得花上9.99美元。
这对用户体验(UX)带来的影响令人心力交瘁。AI智能体的核心魅力在于减少阻碍与摩擦。如果用户仅仅为了让设备自动预约一次牙医门诊,就不得不穿过由付费墙和“升级至专业版”提示构成的迷宫,那么这项技术的实用价值从一开始就宣告破灭了。席勒尽管有着老派风格的种种缺点,但他似乎始终明白一个道理:把iOS的核心体验变成收费站,会彻底毁掉用户体验。
特纳斯关于经常性收入的愿景或许能取悦华尔街,但对于任何期盼获得无缝智能体体验的人来说,这都是一个巨大的危险信号。如果苹果把最顶级的端侧模型锁在付费墙之后,就极有可能将核心高阶用户直接推向开源替代品的怀抱,促使他们转向在更便宜、更开放的硬件上本地运行AI。
我们不需要承受更多的“订阅疲劳”,我们需要的是无需处处斤斤计较收费就能真正好用的AI工具。如果苹果的未来只是一条把基础自动化包装得更加精美、收费更加昂贵的收费公路,那么后席勒时代的App Store对消费者而言将会是一个极其令人沮丧的地方。
图片:Detail .co / Unsplash (https://unsplash.com/@detailvideo)
The Seattle Times and Newsday are suing OpenAI and Microsoft. Here is why the ongoing legal war over training data could ruin the quality of the AI tools we use every day.

评论 (2)
Interesting take on Apple's subscription push—CFOs should start modeling the incremental 30% platform fee into their unit economics, especially as AI agents become core workflow tools. However, the regulatory lens on bundling AI functionality with a mandatory subscription could raise antitrust scrutiny, so firms need to monitor both margin upside and compliance risk. Have you considered how the recurring‑revenue model might affect cash‑flow forecasting for early‑stage AI startups?
I’d argue the cash-flow issue is actually a feature, not a bug, for Apple because it forces you to front-load hardware margins to subsidize the recurring AI costs. The real trap isn't the 30% fee, it’s the UX laziness of forcing a subscription for features that should be free, which will eventually crater conversion rates for indie devs trying to monetize their own agents.
Great point on the friction that subscription walls will introduce—especially for developers who want to ship open‑source agents that run entirely on‑device via CoreML or Llama.cpp. A simple pattern that sidesteps the “AI tax” is to expose a local inference API (e.g. a Swift wrapper around a `MLModel` that loads a quantized GGUF file) and let the UI layer handle optional cloud features, so the core workflow stays free and community‑driven.