A while ago I suggested that by H2 2024 we would begin to see an influx of close-proximity, on-device AI agents. Chipmaker roadmaps pointed in that direction early: dedicated NPU capacity made lower-latency, embedded agentic capability a practical target rather than a distant one.
Now that Microsoft, Google, and Apple have all shown their platform direction, the foundation is clearly in place for AI-assisted consumer creation, discovery, and distribution. What is interesting is not only what these platforms can do, but how quietly they are starting to do it.
From Prompted Services to Embedded Assistance
Most people still experience AI through transactional services such as ChatGPT, Claude, and Copilot search: you ask, it responds. That model requires explicit user initiation and reasonably well-formed intent, much like classic search did.
The next wave is different. In-device services are increasingly contextual, persistent, and proactive. They are less about visible conversations and more about ambient capability running underneath everyday workflows.
Why This Matters
Our digital assets are growing rapidly and becoming more fragmented across apps, clouds, and devices. On-device personal AI has the potential to reduce that friction by helping people create, find, and share what matters without the same manual overhead.
In practice, that means less effort in scenarios like finding a specific memory in a large photo archive, assembling life-event media, planning last-minute travel, or making better purchase decisions. The underlying mechanism — context awareness and inference running persistently in the background — would have sounded implausible five years ago. Today it is already deployed across hundreds of millions of devices.
The Real Trick
The greatest trick may be that the most powerful AI shifts are becoming less visible. These systems can monitor context, infer intent, and deliver value before users explicitly ask. In other words, AI will be everywhere, but not always obvious.
After testing these ecosystems hands-on, the direction is clear: these capabilities are real, useful, and close to mainstream. Many people will soon be using AI deeply, even when they do not describe it as AI usage. That is the real trick: not that AI has become powerful, but that it has become invisible. The most significant capability shift in a generation, and most people will absorb it as just another feature.