An on-device AI layer that pulls tasks out of WhatsApp threads and collects them in one tab
Millions of people already use WhatsApp as their de facto task manager, dropping notes into self-chats and splitting responsibilities across group threads. There is no structured task layer to hold any of it, so critical action items sink into long message histories and users end up running Google Keep, Todoist and Calendar alongside the chat app.
WhatsApp To-Dos detects task-like messages on device and surfaces them in a dedicated tab. Confidence drives the interruption: high-confidence tasks are added silently so the conversation is never broken, while medium-confidence items raise a reviewable suggestion banner. Manual control stays with the user through long-press gestures, alongside reminders, snoozing, marking complete, and assigning a task to someone else in a group chat. Detection runs locally, so no message content leaves the phone by default and end-to-end encryption holds.
Worth stealing
Read this if
PMs adding AI detection to a product where trust is the moat
The transferable bitInterruption is tied to model confidence rather than one behaviour for every prediction, which is the design decision to copy. Running detection on device is the other: it turns a privacy constraint into the feature's reason for existing, which is a far stronger position than bolting privacy language onto a cloud feature after the fact.
Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.