YC Startup EdotEnv: More Reasoning Doesn't Make AI Agents Better Traders

YC S26 startup EdotEnv reports that advanced LLM agents in its real-market reinforcement learning environments gain nothing from higher reasoning effort. When losing money, the agents stop trading instead of adapting. These failure modes wouldn't appear in synthetic benchmarks, where inaction is rarely penalized.