Agents Prompt People: Using Agents for Long-Horizon Financial Work
Arca's AI wealth management agents prompt human advisors rather than replacing their decisions
“agents prompt people more than people prompt agents.”
Arca, an AI-native wealth management firm backed by $48.5M from General Catalyst, has built a human-in-the-loop agent architecture where AI handles analysis and long-horizon task tracking (up to 10 days) but humans retain final decision authority on all trades and money movement. The firm built a custom model harness that auto-selects between Anthropic and OpenAI models, with Claude as primary, and enforces fully deterministic safety restrictions that remove risky decisions from agent control. This is a concrete production example of agentic AI in regulated industries with a clear pattern for safely deploying long-running agents.