Solving false triggers in always-on agents: Why eye contact beats wake words
deebuildsthings · reddit · 2026-08-23
This post details the iteration process for "Autonomous Lamp," a desk robot designed as a companion, focusing on how to solve false triggers caused by environmental noise.
The Evolution:
- V1 (Always Responding): Reacted to every sound (typing, coworkers), creating a noisy experience.
- V2 (Wake Word): Implemented "Hey Lamp." It worked perfectly but felt like operating a device rather than living with a companion.
V3 (Eye Contact):
Realizing people use body language to initiate conversation, the team implemented a system that checks if the user is facing the robot before responding. It uses a 3-second camera buffer to track head orientation. If you look at it and speak, it wakes up. This creates a natural, button-free interaction.
More from coding & agent
- Agent Hack: Instructing to Continue Until All Work is Complete — tristanbob · 2026-08-23
- Using AI to plot long-term performance graphs helps catch subtle regressions — lemire · 2026-08-23
- Agent Safety: Labeling Tool 'Blast Radius' for Risk Control — blaizedsouza · 2026-08-23
- Agent Retry Policy Guide: Configure Per Tool, Not Globally — blaizedsouza · 2026-08-23
- Microsoft releases AI Engineering Coach to audit coding agent sessions — tom_doerr · 2026-08-23
- Firecrawl: Clean Web Scraping for AI Agents — alex_verem · 2026-08-23