Arca, an “AI-native” wealth management firm that recently raised $48.5 million in a round led by General Catalyst with participation from Index Ventures and Venrock, is using AI agents to do most of the background work on client matters, while human advisors make the final decisions. Lead AI engineer Aman Singh told Latent Space that the goal is for AI to guide the process, in his words, “agents prompt people more than people prompt agents.” Agent tasks at the firm can run for up to ten days, with many handoffs and decision points. To manage that, Arca has built an internal memory system that tracks what has happened for each client, what is pending in the next few days, what is planned for the longer term, and which individual is responsible for each action. Arca's agents are limited to analysis and proposing solutions. Trading and any other money movement stay with human advisors, and Singh said the safety restrictions are fully deterministic, which takes those risk decisions out of the agent's hands. The firm uses Claude as its primary model provider alongside other frontier models, and it runs its own harness that selects the model for each task based on internal benchmarks. Singh said owning the harness means Arca can switch providers if needed. Agents write code to query client data inside sandboxes, and that code's output is forced into a strict format and checked by guardrails before it reaches back-end systems. A “meta-agent” analyzes advisor feedback and agent traces to suggest harness changes. Arca also runs checks for known AI security issues such as prompt injection. For Australian businesses and professionals outside wealth management, this changes little day to day. For firms giving personal financial advice, the design pattern is worth a closer look: keep money movement behind human approval, enforce hard limits in code rather than relying on prompts, and keep audit trails. Any Australian firm that sends client data to overseas model providers should check its obligations under the Privacy Act 1988, including the Australian Privacy Principles on cross-border disclosure, and confirm its licensing position for personal advice with ASIC. The article does not address Australian rules, and its figures and descriptions come from company statements made in an interview rather than independent verification.