Product thinking

All-day Apple Watch recording: your AI should understand you better

Xuesheng Guo

This article preserves the product version, test conditions and views at the time of writing. Features, pricing, data handling and download options may have changed; consult the current product page.

In the past couple of days, I've shifted TalkTuck's positioning from an offline iWatch recorder to a personal AI assistant on the phone, and redesigned the interface.

Follow along for access to the beta.

Because I originally built a recording app, I instinctively thought transcription was central. I assumed everyone would check a transcript after recording. In practice, most people open the app, record, leave it, and return when they need to search or organize something.

That suggests I should focus on recording and presenting the content while making the recording itself less prominent. The home screen has changed from a recording list to content summaries. When people reopen the app, they can see material organized by AI.

To find a recording, you can tap search for a keyword match or use AI to find related passages.

In the agent era, using AI is becoming more expensive. At the time of writing, commercial transcription plus a large model costs about 2–3 yuan a day, or 60–90 yuan a month, which may not be good value for many people. I'm exploring a balance between cost and quality: offline models, filtering unusable recordings and reducing token use are all options.

What do you think about an on-device personal AI assistant for your phone? I'd love to hear your thoughts.