24‑hour AI trading bots

A developer spent 24 hours building a four-bot trading setup with Claude AI agents and posted the workflow as a public thread. (anthropic.com) The setup came from Moon Dev, a YouTube creator who publishes trading and coding videos under the handle @moondevonyt. His broader open documentation describes an experimental crypto-trading system with more than 48 specialized agents for research, strategy development, risk checks and execution. (youtube.com) (deepwiki.com) The core idea is simple: use one agent to research a trading idea, another to code a backtest, another to review results, and another to prepare the strategy for live use. Moon Dev labels that loop “Research, Backtest, Implement,” or “Research-Based Inference” in his public materials. (deepwiki.com) (youtube.com) A backtest is a replay of an old market with a new rule set, like testing a chess opening against recorded games before using it in a tournament. Moon Dev’s public repository notes that its RBI agent takes a video, document or prompt, researches the strategy, and hands it off for code generation and testing. (github.com) (deepwiki.com) Claude Code is the engine behind that speed. Anthropic says the tool can read a codebase, edit files across projects, run command-line tools and rerun tests, which lets one person supervise several coding tasks in parallel instead of writing every script by hand. (anthropic.com) (github.com) Anthropic’s training materials now include lessons on “agent skills” and “subagents,” which are reusable instructions and delegated helper agents inside Claude Code. That matches the kind of multi-agent workflow Moon Dev describes in his trading posts and documentation. (anthropic.skilljar.com) (deepwiki.com) The pitch is not that the model predicts markets on its own. The pitch is that artificial intelligence compresses the engineering work around trading ideas: collecting rules, writing test code, checking edge cases and packaging a bot fast enough to try several versions in a day. (anthropic.com) (deepwiki.com) That speed does not settle the harder question of whether any strategy will hold up with real money. Moon Dev’s public materials describe circuit breakers, loss limits and position controls, and his recent video descriptions warn viewers that he cannot guarantee profits from any bot or strategy. (deepwiki.com) (youtube.com) The thread’s real claim is about workflow, not verified returns: one person with agentic coding tools can now assemble, test and iterate trading systems in hours instead of weeks. That is the shift Moon Dev keeps pointing to as he turns trading research into a supervised swarm of bots. (anthropic.com) (deepwiki.com)