Arca says agents handle wealth work, humans keep final decisions
Original title: Agents Prompt People: Using Agents for Long-Horizon Financial Work
Arca raised $48.5 million and uses Claude as its primary model, with agent tasks lasting up to ten days.
In a Latent Space interview, Arca lead AI engineer Aman Singh described an AI-native wealth firm where agents do background analysis and propose options, while human advisors make final decisions and handle trading and money movement. Tasks can run up to ten days with many handoffs, so Arca tracks client state in an internal memory system. The firm uses Anthropic and OpenAI models, mainly Claude, through its own harness that routes work using internal benchmarks and checks for prompt injection.
- General Catalyst led a $48.5 million round with Index Ventures and Venrock.
- Singh said safety restrictions are fully deterministic and remove risky decisions from agents.
- Agents may query data through sandboxed code that is rewritten into a strict format.
- A meta-agent reviews advisor feedback and traces to suggest harness changes.
- Organizations
- ArcaGeneral CatalystIndex VenturesVenrock
- AI models
- Claude
From the original article
3 min read at latent.space
Arca's head of AI discusses persistent memory, the benefits of a custom harness, and agentic workflows that span days.
By Richard MacManus · latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.