Your agent's paper-trading bot runs for weeks after the tool call returns — how much state should it see?
QuanTradin · reddit · 2026-10-19
The author built Quantradin: an AI writes a trading rule, a separate engine backtests it without look-ahead, and surviving rules keep running as paper bots the AI can't touch (opening positions requires a deliberately issued key; narrower keys can only close/cancel/pause; free to use).
The hard design problem: the bot keeps trading for days or weeks after the conversation that started it is gone. When a new session asks "how's my bot doing," what should the agent get? Three options, each with tradeoffs:
- A status tool (positions, recent fills, rule violations) — risk: the agent sees three red days and "fixes" a fine rule, overfitting on live noise
- An event log since its last check — reads deltas, same overfitting worry
- Nothing; humans read a dashboard and the agent acts only when asked — but then the agent is blind, defeating the point
The author asks anyone running agents against long-running things (jobs, deploys, monitors): do you let the model see running state, or only a summary on request? Did more visibility make it act worse?
More from coding & agent
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23