Latent Space: The AI Engineer Podcast · Oct 31, 2026 · 26m · 18m saved
With Aman Singh, Lead AI Engineer at Arca
Photo via Latent Space
Arca's AI agents can spend up to 10 days working out how a client should raise cash for an unexpected tax bill, and at the end of it they still cannot move a dollar. Aman Singh, lead AI engineer at the AI-native wealth management firm Arca, explained the design to Richard MacManus on Latent Space: The AI Engineer Podcast.
Most financial firms add AI tools that staff have to prompt. At Arca the direction runs the other way: agents work around the clock and bring analysis, deadlines and recommendations to the human advisors.
"The role of AI at Arca is not to replace the advisor. It is to replace all of the work around the advisor." — Aman Singh
Aman leads the team that built Arca's AI platform. The firm recently raised $48.5 million in a round led by General Catalyst, with Index Ventures and Venrock taking part, and its clients range from families buying a first home to tech executives planning an early retirement.
The full interview is covered here so you can skip it. 26 minutes of audio, 8 minutes of reading.
Here are the 10 principles that matter.
Key Takeaways
At Arca, agents prompt the advisors, rather than advisors prompting agents
Arca built its own model-agnostic harness instead of using a lab's managed-agent product
Its own benchmarks, not public leaderboards, decide which model gets which task
Agents can analyze money but cannot move it: trades and transfers need a person with authority
Every safety rule is deterministic code, which takes the decision away from the model
Every action, by an agent or a person, leaves a change log, so compliance gets an audit trail by default
A nightly meta-agent proposes fixes to the harness, but an engineer reviews every prompt change
Arca's agents found fees a prospect's advisor had not disclosed, and that the advisor was not a fiduciary
1. Agents Prompt the Advisors
Arca is a wealth management firm that runs most of its operations on agents. It offers financial planning, tax and estate planning, and insurance, with human advisors as the client-facing layer and the final decision-makers.
Arca says it is the firm itself, not a software vendor selling AI to firms
"We're actually not a SaaS company. We are the firm ourselves. Agents prompt people more than people prompt agents, and it is a proactive kind of service that is running for them twenty-four seven." — Aman Singh
The difference is who starts the work. An advisor at Arca does not have to ask an agent for anything. Agents raise the analysis, the deadline or the recommendation first.
