Hands on with Facebook's virtual assistant ‘M’: promising Siri competitor but often slow to respond to complex questions
Owen Williams / The Next Web :
Context & Ripple Effects
Facebook is testing M, an AI-based assistant with human supervision inside Messenger that can complete tasks like purchases on your behalf (the original test announcement). Wired has already explained how the AI layer works (how M's artificial intelligence works); this hands-on from Owen Williams is the first real-usage read on whether it holds up.
The verdict splits: M looks like a genuine Siri competitor on capability, but complex questions often come back slowly — a preview of the tension between quality and responsiveness that The Verge's later hands-on and David Marcus's comments on scaling would pick up in November.
First-order effects
- Users in the Messenger test get an assistant that can handle real tasks beyond Siri's scope, but pay for it in latency when queries get complicated.
- Siri and Cortana face an early benchmark set not by raw AI accuracy but by what a human-supervised system completes end-to-end.
Second-order effects
- Every complex request M answers routes through human labor, so Facebook's cost per task scales linearly — the exact scaling problem Marcus was pressed on in the related coverage.
- Apple, Microsoft, and other assistant makers are pushed to justify why their fully automated assistants can't complete purchases and errands, shifting the competitive axis toward task completion.
Third-order effects
- If the human-in-the-loop pattern holds across the industry, virtual assistants bifurcate into fast-but-limited AI-only tiers and slower high-touch hybrid services, forcing providers to price and staff accordingly.
- Assistant competition moves toward whoever owns the messaging surface where tasks happen — which is why Facebook embedding M in Messenger rather than shipping a standalone app matters structurally.
The trend: Virtual assistants are splitting into pure-AI responders and human-supervised task completers, with response speed versus capability becoming the trade-off that defines each tier.