On 781 Agent-Run Trading Bots, the Only Ones With Risk Limits Set Them Themselves
QuanTradin · reddit · 2026-09-04
A trading-platform author found that across 781 agent-deployed bots on 16 desks, only 18 bots had risk limits — all written by the agents themselves into strategy specs, while zero humans ever used the risk settings page (including owners of 169 bots). Bots with limits lost $21,414; the 763 without lost $1,330,205 (paper money). He now debates adding an agent-facing, tighten-only risk ratchet.
More from coding & agent
- Background computer use for Codex on Windows confirmed incoming by OpenAI staff — pvncher · 2026-09-04
- Building production agents? The hard part is idempotent writes across apps, not model reasoning — anishfish · 2026-09-04
- AI made code cheap, not engineering: why judgment matters more than ever — _jaydeepkarale · 2026-09-04
- xAI to Host Grok Bot Galaxy: Three-Day Event Teaching Users to Build Custom AI Agents — Kyrannio · 2026-09-04
- Garry Tan: His Personal AGI Is a Folder of Markdown Files, Shipping 400x His 2013 Output — garrytan · 2026-09-04
- Hivemind launches persistent memory infra for agents: unlimited context, shared swarm mind — Kyrannio · 2026-09-04