Physix FrontierPhysix Frontier · News Briefing Card (Hacker News · Oct 20, 2026)
Jev Launches Decision Model for Proactive AI Agents in Group Chats
KEY FACTS
- Jev proposes replacing the @mention mechanism with a small decision model that determines when an AI agent should speak in a group chat.
- The decision model does not generate text; instead, it returns calibrated probabilities for each message with millisecond-level latency.
- Its cost is about $0.04 per million input tokens, allowing it to score every message in busy channels.
- Runtype's Slack judge mode has entered preview, supporting a shadow mode that logs first and speaks later.
- Jev supports zero data retention, enabling per-message classification in sensitive channels.
KEY DATA
about $0.04 per million tokensDecision model input cost
PHYSIX OBSERVATION
Letting agents decide for themselves when to speak is closer to real collaboration than forcing people to remember which bot to @mention. The decision model drives costs down to negligible, meaning the bottleneck for group chat AI shifts from budget to strategy design. But once proactive speaking misjudges, noise is amplified too, and shadow mode is a necessary safety net.
Source: Hacker News report