Latent Space: The AI Engineer Podcast

ARCA puts agents to work on long-horizon wealth management

Using Agents for Long-Horizon Financial Work — Aman Singh, Arca

The episode shows how financial firms can use persistent, auditable agents without surrendering control of trading, money movement, or client decisions.

3 key takeaways
  1. 1ARCA assigns models by task while hiding orchestration complexity from advisors and preserving a model-agnostic system.
  2. 2Agents maintain context across multiday portfolio workflows, but humans control liquidation, trading, communication, and other sensitive actions.
  3. 3Audit trails, curated operational knowledge, and staff feedback turn everyday financial work into measurable system improvement.

Don't miss

Singh explains how ARCA handles a tax bill: agents analyze the portfolio and propose a cash-raising plan, while humans control every sensitive action.

The brief

Aman Singh describes ARCA’s AI-native wealth-management model, where agents support human advisors across investment, tax, insurance, estate-planning, and client-service work.

ARCA’s internal harness routes tasks across model tiers, keeping model selection abstracted from advisors while matching computational effort to the work’s complexity.

In a tax-bill scenario, agents analyze a portfolio and propose how to raise cash; humans retain control of liquidation, money movement, and client communication.

The hardest workflows can last ten days as emails and calls change the facts, so persistent memory, verified writes, deterministic restrictions, and audit logs keep plans safe.

ARCA does not let agents trade autonomously: staff edits and cancellations become evaluation data, while curated operational knowledge gives agents current guidance and citations.

Weekly email

State of the Pod, every Monday

The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.

Listen to the full episode and explore every guest, topic, and moment on PodLume.