Ray-Ban Meta glasses hands-on: AI features coming in April, triggered by “Hey, Meta”, are sometimes impressive and helpful, but the AI often gets things wrong
Brian X. Chen, left, and Mike Isaac, reporters for The New York Times, trying out Meta's new Ray-Ban smart glasses.Aaron Wojack for The New York Times
Context & Ripple Effects
Meta had already put multimodal AI into the glasses through a US early-access beta, extending a product whose camera and audio hardware had improved faster than its AI usefulness. This hands-on test is an early check on whether voice-triggered visual assistance can become a dependable reason to wear the device.
The mixed results matter because the glasses place an AI assistant in a low-friction, camera-equipped form factor: useful answers can feel immediate, while wrong answers are especially conspicuous when users rely on them in the moment.
First-order effects
- Ray-Ban Meta users scheduled to receive the April features gain hands-free visual queries through “Hey, Meta,” but must treat outputs as assistive rather than authoritative given the observed errors.
- Meta gets evidence that multimodal interaction on glasses can be compelling in specific moments, alongside a clear reliability gap that can limit repeat use and trust.
Second-order effects
- The rollout raises the product bar for smart-glasses rivals: camera, audio, and voice control alone are less differentiated when AI can interpret what the wearer sees, but accuracy becomes a central competitive metric.
- Use cases such as object identification, sign translation, and social posting will depend on reliable results; subsequent early-access capabilities for those tasks make the quality of the underlying visual AI more consequential.
Third-order effects
- If reliability improves, smart glasses could shift AI interaction from deliberate screen-based prompts toward ambient, sensor-assisted help; if it does not, the category may remain a novelty despite capable hardware.
- The key structural test is whether companies can earn trust for always-available visual AI without encouraging users to over-rely on error-prone answers in everyday settings.
The trend: This is an early data point in the push to distribute multimodal AI through wearable, sensor-native devices rather than standalone apps.