Agents Prompt People: Using Agents for Long-Horizon Financial Work

At Arca, a wealth management firm, AI agents analyse accounts and propose plans, but fixed rules keep them away from client money and human advisers make every trade.

Richard MacManus of Latent Space interviewed Aman Singh, lead AI engineer at Arca, a wealth management firm where AI agents do most of the background work. Singh described agent tasks that last up to ten days, with many handoffs between people. The agents analyse and propose, while human financial advisers make the final decisions and handle every trade.

• The agents keep a memory for each client of what has happened, what is due next and which person owns each action.
• Fixed rules, not the agent, decide which tools and tasks an agent may run, and anything that puts client money at risk is excluded.
• Code that an agent writes to query client data is translated into a strict format and checked before it reaches Arca's systems.
• Feedback from advisers and records of agent runs go to a second AI agent, which suggests changes to the software that runs the agents.

A team putting AI agents into regulated work can keep each risky action out of the agents' reach by rule rather than by instruction. The account is a short interview summary and gives no figures on results.

Derived by Working Surface from the article; more in the Playbook. Source line: agents prompt people more than people prompt agents.

1 November 2026: Teams putting AI agents to work decide in advance which steps a person must still take, and that person carries more review and a greater duty to push back.