Autonomous Trading Agents
Autonomous trading agents are AI agents given market data, brokerage access, strategy instructions, and recurring execution authority to research, trade, rebalance, and review portfolios with limited human intervention [src-010, src-086].
Key points
- In [src-010], Nate Herk demonstrates the pattern with a Claude Code cloud-routine agent connected to Alpaca Markets and market-research tools.
- In [src-086], Nate and Salmon run a 30-day real-money experiment where separate OpenClaw agents each manage $10,000 accounts under different prompts and strategies.
- The experiment used human monitoring but no mid-run human strategy changes, turning the setup into a sandbox for evaluating agent behavior rather than a manual trading challenge [src-086].
- Nate's bot "Bull" used Alpaca Markets, Telegram interaction, broad wealth-adviser prompting, and an emergent hybrid momentum/options strategy with cash and position-size constraints [src-086].
- Salmon's bot used an OpenClaw setup with selected signal sources, cron-like checks during market hours, Discord monitoring, and portfolio rebalancing based on signals and news [src-086].
- Both agents beat a sharply falling S&P 500 baseline during the test period, but the result is not evidence of a durable trading edge; it is mainly evidence about agent authority, monitoring, memory, and risk controls [src-086].
- The safety lesson is that real-money agents need explicit permissions, account isolation, budget limits, order constraints, audit logs, and human override paths before they are trusted with higher stakes [src-086].
Related entities
Related concepts
- Claude Code Cloud Routines
- Agent Security Boundaries
- Agent Budget Controls
- Stateless Agent Memory Pattern