AI is a technology not a product
468 points
• 1 day ago
• Article
Link
John Gruber 对 Steven Levy 的观点提出了异议。 Levy 认为 Apple 的下一任 CEO 需要推出一款"杀手级 AI 产品"，像 iPhone 那样定义 AI 时代。 Gruber 认为 Levy 被 AI 炒作带偏了；Apple 一向不是单纯发布底层技术，而是打造出色的产品和体验。 iPod 不是关于 MP3 或微型硬盘，而是关于听音乐。 iPhone 定义了移动时代，但 Apple 不必在那个时代衍生出的每个市场（比如社交媒体）都称霸，才能保持成功。
Gruber 对 Levy 预测 AI 代理会很快取代 iPhone 生态、主动处理叫车等事务持怀疑态度。他把这种预测称为"被炒作冲昏头脑的幻想"，质疑这样的系统如何运行、依赖什么硬件，以及人们是否会觉得它是便利而非令人不适。他怀疑到 2030 年人们会放弃用手机打车，无论是语音还是传统点按。
他认为用手表、耳机或眼镜等更小的设备来替代手机进行 AI 交互并不合理，因为这些设备很可能仍需与随身携带的手机配对。 Apple 在制造小型个人计算设备方面已是世界一流，但手机在相机、屏幕和日常使用中仍居核心地位。放弃手机转而使用独立小型设备的想法难以成立。
Gruber 的核心观点是：AI 是一项技术，而不是独立的产品类别。他将其比作无线网络——Wi‑Fi 、蜂窝和蓝牙已渗透到 Apple 的所有产品中，但它们本身并不是某个单一的"杀手级产品"。 AI 同样会融入所有产品，而不会靠某一台设备来定义时代。 Apple 无法忽视 AI，因为它无处不在，但这并不意味着必须做出一款革命性的 AI 产品才能保持竞争力。
John Gruber pushes back against Steven Levy's argument that Apple's next CEO needs to launch a "killer AI product" to define the AI age the way the iPhone defined the mobile era. Gruber finds Levy's reasoning to be caught up in AI hype, arguing that Apple's philosophy has never been about shipping raw technology but about shipping great products and experiences. The iPod wasn't about MP3 files or tiny hard drives, it was about music. The iPhone defined mobile, but Apple doesn't need to dominate every market that era opened up, like social media, to remain successful.
Gruber is particularly skeptical of Levy's prediction that AI agents will soon replace the iPhone ecosystem by proactively handling tasks like summoning ride-shares without any user friction. He calls this a "fever dream high-on-the-hype fantasy," questioning the practicalities of how such a system would work, what hardware would support it, and whether people would even find it desirable rather than creepy. He doubts that by 2030, people will be using anything other than their phones to hail rides, whether through voice commands or traditional tapping.
Gruber argues that the idea of smaller devices like watches, earbuds, or glasses replacing phones for AI interactions doesn't make sense, since those devices would likely still pair with the phone you're carrying anyway. He points out that Apple is already the best in the world at making smaller personal computing devices, yet the phone remains central for cameras, screens, and general use. The notion that we'd abandon the phone for independent smaller devices seems implausible to him.
The core of Gruber's argument is that AI is technology, not a product category. He compares it to wireless networking, which is now pervasive across all Apple devices but isn't itself a standalone "killer product." Just as Wi-Fi, cellular, and Bluetooth permeate everything Apple makes without being a single flagship offering, AI will similarly be integrated into all products rather than being one defining device. Apple can't ignore AI because it's pervasive, but that doesn't mean it needs to create one revolutionary AI product to stay relevant.
207 comments • Comments Link
对苹果来说，最有价值的人工智能应用，是让 Siri 真正变得好用：用户可以用自然语言安排日历、控制应用、创建可重复使用的快捷方式，而无需遵循特定语法。关键是要满足用户的实用需求，而不是把人工智能当作炫技的目标。
科技公司已经偏离了以用户价值为中心的路线，反而把整合得不佳的 AI 功能强行塞进产品里，却没有考虑这些功能是否真正改善了用户工作流程。 Claude Code 被认为是一个良好范例，因为其开发者同时也是产品的使用者，能更贴近实际需求。
目前许多 AI 应用存在问题：模糊搜索和反向词典这样工具性的功能本身就很有用，无需生成式 AI 就能实现；而 OCR 、机器翻译等应用在引入便利的同时也带来了新的故障模式和滥用风险。
确定性搜索工具允许用户用加权参数构建精确查询，在信息检索方面远比生成式 AI 更可靠。 PubMed 是复杂搜索逻辑的典型例子，显示出人工智能无法替代某些严谨检索场景的局限。
理想的软件策略是在内部利用 AI 进行优化，同时向有明确需求的用户开放更多参数，并引导那些不确定自己需要什么的用户——而不是把控制权交给自治系统，从而剥夺用户决策权。
像 Siri 这样的语音界面存在根本性的体验问题：隐私保护不足、误触频繁，以及在处理大量信息时的局限。这说明语音交互应以刻意激活为主，而不是一直监听的唤醒词。
苹果历来从客户体验出发、反向驱动技术开发，这意味着 AI 应当作为无形的基础设施来增强产品功能，而不是把"AI"本身作为噱头来营销。
让 AI 完全代替人为日常事务（比如叫车）的前景，遭到不少用户反对：他们并不认为日常琐事是令人厌烦到必须自动化的对象。
也有人为代理式自动化辩护，认为它在应对应用中的"暗黑模式"和复杂决策时，可能比手动操作更能代表用户利益，尤其在与操控性界面打交道时。
在可预见的未来，智能手机仍将是 AI 的主要载体。只要苹果坚持"让技术为用户让路"的设计理念，就能保持其竞争优势。
苹果面临的创新困境在于，iPhone 的高盈利性可能抑制公司去追求更激进的外形变革，例如最终可能取代手机、成为核心个人设备的高级智能手表。
AI 本身是中性的，但在企业利益、法律责任忧虑和资源集中驱动下出现了问题。解决之道在于推广高效的本地模型和开源开发，从而实现技术的民主化。
能进行网络搜索的本地模型对用户更有价值，因为真正的价值往往来自于搜索与信息获取能力，而不仅仅是生成文字的能力。
小型企业已经从大型语言模型中获益，例如在建站方面的帮助，让缺乏技术能力或负担不起网页设计师的人能开展在线业务。但也有人质疑，相较于现有模板解决方案，这是否真能构成实质性进步。
讨论暴露出一个根本矛盾：AI 是应被构建为实用的基础设施，还是应被设计为自治代理？多数参与者倾向于将 AI 无形地整合进现有工作流，而不是替代用户控制的系统。当前很多 AI 实现忽视了用户的真实需求，更多偏向技术演示。苹果的优势在于始终以客户体验为出发点，而非被技术驱动。同时，讨论也强调了企业对 AI 系统控制的风险，开源和本地模型是对集中式系统的重要制衡，因为集中系统不可避免地反映企业利益与偏见。 • The most valuable AI implementation for Apple would be making Siri actually functional, allowing natural language commands for calendar events, app control, and reusable shortcuts without requiring specific syntax, with the key insight that users want practical utility rather than AI for its own sake.
• Tech companies have lost focus on user value, instead forcing poorly integrated AI features into products while ignoring whether they actually improve user workflows, with Claude Code cited as a positive example because its developers are also users.
• Many current AI applications are problematic, with only fuzzy search and reverse dictionary functions being unalloyed goods that don't require generative AI, while other uses like OCR and machine translation introduce new failure modes and enable exploitation.
• Deterministic search tools that allow users to craft precise queries with weighted parameters remain far more powerful than AI for information retrieval, with PubMed cited as an example of sophisticated search logic that AI cannot replace.
• The ideal software approach combines AI internally to refine and expose more parameters to users who know what they want, while guiding those who don't, rather than removing user control and decision-making to autonomous systems.
• Voice interfaces like Siri have fundamental UX problems including lack of privacy, accidental activations, and difficulty with dense information, suggesting they should require deliberate activation rather than always-listening wake words.
• Apple's historical approach of working backward from customer experience rather than technology means AI should be invisible infrastructure that enhances products rather than being marketed as a feature itself.
• The vision of fully autonomous AI agents handling tasks like booking rides is criticized as undesirable by many users who enjoy the small decisions of daily life and don't view them as burdensome drudgery requiring automation.
• Some defend agent-style automation as potentially beneficial for handling dark patterns and complex decisions in apps, representing user interests more effectively than manual interaction with manipulative interfaces.
• The smartphone form factor will persist as the primary AI conduit for the foreseeable future, keeping Apple in a strong position as long as they maintain their design philosophy of making technology that gets out of the way.
• Apple's innovator's dilemma with the iPhone's massive profitability may prevent them from pursuing more radical form factors like advanced smartwatches that could eventually replace phones as the central personal device.
• AI itself is neutral technology that becomes problematic through implementation choices driven by corporate interests, liability concerns, and resource concentration, with the solution being democratization through efficient local models and open-source development.
• Local AI models capable of web searching would be more valuable to users than standalone frontier models, as the real value lies in search and information access rather than generative capabilities alone.
• Small businesses have materially benefited from LLMs for website creation, enabling those without technical skills or budgets for web designers to establish online presences, though some question whether this represents genuine improvement over existing template solutions.
The discussion reveals a fundamental tension between AI as practical infrastructure versus AI as autonomous agent, with most participants favoring invisible integration that enhances existing workflows rather than systems that remove user control. There's strong consensus that current AI implementations often ignore actual user needs in favor of technological demonstration, with Apple's potential advantage lying in their historical focus on customer experience over technology-first approaches. The conversation also highlights concerns about corporate control of AI systems and the importance of open-source, local models as a counterbalance to centralized implementations that inevitably reflect corporate interests and biases.