Using Agents for Long-Horizon Financial Work — Aman Singh, Arca
Arca, an AI-native wealth firm, uses agents for everything around the human advisor, keeping agents proactive and deterministically guarded, and owning its harness, memory and knowledge instead of renting them from labs.
Watch on YouTube ↗Charlie's summary
- ▶ 1:29
AI should replace the work around the advisor, not the advisor
Aman says Arca's AI is not meant to replace the advisor but “all of the work around the advisor.” Clients never talk to AI directly. Behind each client, agents keep thinking about what the market is doing and how it affects them, so service is more proactive and responsive. It is still delivered through a human advisor's empathy and care.
- ▶ 2:33
Choose models with internal benchmarks and keep your own model-agnostic harness
Arca has moved from Gemini to OpenAI to Anthropic, with Claude as its primary model. Aman says internal benchmarks drive the choice. A model that looked great on external benchmarks was too slow and costly in their harness. Tasks are tiered by complexity: tax analysis gets the heaviest models, background work gets lower tiers. Advisors never see this routing. The goal is not to depend on one provider or lab harness.
- ▶ 5:36
Agents do the analysis, but humans must approve anything that moves money or talks to clients
In a 10-day tax-bill example, agents fully automate the analysis of where a client's money sits and the best way to raise $100,000. The custodian offers APIs for execution, but Arca keeps strict human-in-the-loop guardrails on moving money, liquidating assets, and communicating directly with clients. Agents hand a plan to the right person to review and adjust. They do not trade autonomously.
- ▶ 7:39
Long-running tasks survive through persistent client memory and a nightly dreaming reflection
Replies to an email thread go to the same agent. A fresh signal, like a phone call, wakes a new agent. It catches up from per-client memory: what is happening, who owns what, what is next, and what is tracked long term. A dreaming process reflects on the client's situation and updates that memory. That lets agents pick up a 10-day task and pull in the right person at the right step.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.