Hands-on: Facebook M, powered by humans, excels at carrying out real-world tasks, unlike pure AI virtual assistants like Siri or Cortana
Hands-on with Facebook M: the virtual assistant with a (real) human touch — There's something eerie about realizing you live in a time people once considered a sci-fi fantasy.
Context & Ripple Effects
Facebook M began as an experiment announced in August, when the company said it was testing an AI-based assistant supervised by humans inside Messenger, capable of completing purchases on a user's behalf. An early hands-on found it promising but often slow on complex requests, so the open question was whether the hybrid design could actually deliver.
This new hands-on answers that more favorably: M's human operators let it carry out real-world tasks that pure-AI assistants like Siri and Cortana cannot complete end-to-end. The trade-off surfaces in David Marcus's subsequent remarks on how humans work alongside AI automation inside M — and whether such a service can scale beyond its current form.
First-order effects
- M users get completed errands — purchases and other real-world tasks executed on their behalf — while Siri and Cortana users are left issuing voice commands that stop short of execution.
- Facebook carries the operating cost of human operators behind every M session, making each interaction expensive relative to fully automated rivals.
Second-order effects
- Siri, already assessed in related coverage as having fallen behind Google Assistant and Alexa amid management dysfunction, faces a capability benchmark set by a rival whose assistant closes transactions rather than just answering queries.
- Cortana and other pure-automation assistants face pressure to either accept a functional gap or adopt their own human fallback layer, raising labor costs across the category.
Third-order effects
- If the pattern holds, assistants bifurcate into cheap, scalable voice interfaces and premium execution services where humans guarantee outcomes — with the unresolved scaling problem deciding which model wins the mainstream.
- Messaging platforms, rather than phone operating systems, emerge as the distribution layer for task-completing assistants, shifting the battleground from device defaults to chat apps.
The trend: Virtual assistants are splitting into pure-automation responders and human-supervised agents that actually execute tasks, with Messenger positioning itself as the latter's proving ground.