Facebook tests context-based actions in Messenger via its “M” assistant, such as sending location or stickers
Facebook's “M” virtual assistant hasn't been rolled out to all that many people, but its interactions with a limited user base have helped train Facebook's artificial …
Context & Ripple Effects
M started as a hybrid experiment: an AI assistant supervised by humans inside Messenger, descended from the earlier Moneypenny project that routed requests to real people for research and ordering. Hands-on coverage found that human backing let M complete real-world tasks where pure-AI rivals like Siri and Cortana fell short.
This test marks the pivot point in that arc: instead of executing tasks on command, M begins proactively suggesting context-based actions — sending location, stickers — to a small user base whose interactions train Facebook's models. It is the bridge between the human-powered concierge and the fully automated assistant.
First-order effects
- The limited group of M users gets proactive, in-conversation suggestions for actions like sharing location or stickers, changing M from a task executor they summon into a layer that anticipates within chats.
- Every accepted or ignored suggestion feeds Facebook's training data, letting the company tune the assistant on real Messenger behavior before any wider release.
Second-order effects
- A working suggestion engine gives Facebook the confidence to open M beyond the test cohort — which materialized months later when Messenger rolled M out to all US users with money-sending, location-sharing, and planning built in.
Third-order effects
- The endgame visible in the corpus is the standalone concierge's death: by early 2018 Facebook was [[a:1161277|sunsetting M as a concierge bot while keeping its context-based suggestions alive inside Messenger]], confirming that the durable product was the embedded suggestion layer, not the human-backed service.
The trend: Virtual assistants are migrating from standalone, human-supervised concierge services toward embedded, AI-trained suggestion layers that live inside existing messaging apps.