WWDC 26 补上了 AI,但离真正的 AI 助手还差什么?| S10E15
Summary
This What’s Next|科技早知道 episode interviews Dong Hongguang / 董宏光, founder and CEO of Guangfan Technology / 光帆科技, about whether WWDC 26, Apple Intelligence, and Siri have brought Apple back into the AI assistant race. The source argues that Apple has made a meaningful catch-up move through system integration, device ecosystem, and privacy-oriented edge-cloud design, but still lacks the model strength, AI-native application layer, ecosystem permission model, and multimodal interaction needed for a real personal assistant. Its distinctive contribution is a Wearable AI Assistant thesis: earbuds, watches, and other accepted wearables may be better AI assistant carriers than phones when the assistant needs always-on sensing, voice interaction, physical-world context, and service execution.
Key Claims
- WWDC 26 is framed as “catch-up” rather than a new interaction paradigm: Apple improves the connection between Apple Intelligence and Siri, but does not yet show an assistant that can reliably complete multi-step everyday tasks.
- Apple’s advantage is system-level integration: it can search across apps, draft messages or emails, call cameras and sensors, and coordinate device-level context more naturally than a standalone app.
- Apple’s likely route is ecosystem orchestration, where the operating system sets standards and third-party apps adapt, but the source expects this to take time because platforms and apps must negotiate permissions, traffic, and business value.
- Siri remains strategically important because a platform-native assistant could become the user’s daily entry point, but the source says Apple’s current assistant still looks incomplete when compared with the ideal of ordering, buying, booking, or coordinating work directly.
- Apple is described as cautious because it is a large company with high downside for mistakes, a privacy-centered brand position, and a relatively slow ecosystem-development path.
- The source says Apple has model limitations: partnerships with OpenAI or Gemini can help, but do not automatically give Apple a differentiated assistant ceiling.
- AI hardware failures such as Rabbit R1 are interpreted as software-infrastructure failures, not proof that AI hardware is impossible. Useful hardware needs sensors, cloud and edge compute, AI-native applications, and new multimodal interaction patterns.
- Google is treated as a stronger full-chain AI player because it spans models, TPU/cloud infrastructure, internet services, and Pixel-like hardware, while Apple has stronger devices and system integration but weaker model depth.
- A personal assistant needs both online behavior context and physical-world context; data locked inside super apps limits what any OS-level assistant can understand or do.
- The source describes a new AI OS as a cross-device layer combining hardware scheduling, edge-cloud compute, Agentic Workflow, AI Skills, Model Context Protocol, and multimodal interaction.
- Guangfan Technology / 光帆科技 chooses earbuds and watches because they are already socially accepted, can be worn all day, and can add sensors for low-friction offline context.
- The source argues that phones are not always the best AI assistant carrier: a phone may be in a bag or another room, and users often do not want to stop, unlock, and operate it for small tasks.
- Always-on assistant value appears in moments such as biking while ordering coffee, museum explanation, object lookup, life advice, and reminders that depend on body-proximate sensing and immediate voice response.
- AI glasses are treated as promising but premature because weight, battery life, prescription needs, indoor-outdoor switching, and accidental interaction still limit daily use.
- Guangfan Technology / 光帆科技’s service approach is to connect directly to cloud services and agent-callable capabilities rather than needing control over every phone app interface.
- The source treats permissions as an unresolved AI-native design problem: old software infrastructure often swings between no access and full access, while assistants need finer confirmation and autonomy boundaries.
- The three-year forecast is optimistic but not complete adoption: Dong Hongguang / 董宏光 expects personal AI assistants to become a consensus direction, while phones, PCs, wearables, and home devices keep separate roles.
Key Quotes
“补课” — the episode’s shorthand for Apple’s WWDC AI posture.
“手机不是最佳载体” — the source’s wearable-assistant counterpoint.
“打通所有生态” — the ideal-state assistant requirement.
Connections
- What’s Next|科技早知道 — show context for this S10E15 AI assistant and wearable-hardware discussion.
- Dong Hongguang / 董宏光 and Guangfan Technology / 光帆科技 — guest and company anchoring the wearable AI assistant thesis.
- Apple, Apple Intelligence, Siri, Google, Gemini, and OpenAI — platform and model-company comparison.
- Wearable AI Assistant, AI Plus Terminals, and Smartphone AI Hub — terminal-form-factor debate between phones and body-worn devices.
- AI Assistant Service Entry, Agent Permission Boundaries, OS-Level Context, and Proactive Agents — assistant capabilities that require context, service execution, and permission design.
- Edge-Cloud AI Boundary, Agentic Workflow, AI Skills, Model Context Protocol, and Agent-Facing Interfaces — implementation layers the source says are needed beyond raw model capability.
- [[RayBanSmartGlasses|Ray-Ban smart glasses]] — adjacent wearable example, treated as promising but not yet frictionless enough for all-day AI assistance.
Contradictions
- No direct factual contradiction found. The source creates a strategic tension with AI 时代的超级入口还是手机吗?| S10E17 and Smartphone AI Hub: S10E17 argues phones remain the central AI hub because they combine context, display, compute, identity, and services; this S10E15 source argues wearables are stronger when the assistant needs constant perception, hands-free response, and body-proximate context. The two views can coexist if phones remain the hub while wearables become the always-on sensing and interaction edge.