智能体提示人类：用智能体处理长周期金融工作 — Latent Space

本月在 AI Engineer New York 活动上，我们与 Arca 的首席 AI 工程师 Aman Singh 进行了交流。Arca 是一家提供财富管理服务的“AI 原生”公司。At AI Engineer New York this month, we sat down with Aman Singh, the lead AI Engineer at Arca, an “AI-native” company providing wealth management services.

我们发现，尽管智能体承担了公司大部分幕后工作，但仍设有防护机制，确保最终决策由人类顾问做出。目标是让 AI 为流程提供指导；正如 Singh 恰如其分地所说：“智能体提示人的次数，比人提示智能体的次数更多。”We discovered that while agents do most of the background work in the firm, there are guardrails in place to ensure that human advisors are making the final decisions. The goal is for AI to guide the process; or as Singh nicely put it, “agents prompt people more than people prompt agents.”

Arca 最近在一轮融资中筹集了 4850 万美元，本轮融资由 General Catalyst 领投，Index Ventures 和 Venrock 参投。Arca recently raised $48.5 Million in a round led by General Catalyst, with participation from Index Ventures and Venrock.

辛格解释说，在金融服务中使用 AI 的特点之一，是这类工作通常具有较长的时间跨度——尽管并非持续运行很久。在他的 AIE 演讲中，辛格提到，Arca 的智能体任务最长可能需要十天。“整个过程非常断断续续，中间涉及许多参与者、许多决策点和多次交接，”他说。One of the features of using AI in financial services, explained Singh, is that it’s typically long-horizon work — although not long-running. In his AIE presentation, Singh noted that agent tasks at Arca can take up to ten days. “It's very stop-start, there's many actors in between, many decision points, many handoffs,” he said.

因此，Arca 开发了一套内部记忆系统。For this reason, Arca has developed an internal memory system.

因此，我们为每位客户都建立了这样的持久化记录，”辛格在我们的播客讨论中说道。“目前为止发生了什么？接下来我们要跟踪什么？接下来几天内需要立即完成什么，[以及] 长期来看我们要跟踪什么？每项行动由谁负责？所有这些都由我们的记忆系统进行跟踪，而我们的智能体会为这些系统提供信息。“So we have this persistence for every client,” Singh said in our podcast discussion. “What has happened so far? What are we tracking for next? What needs to happen immediately in the next couple of days [and] what are we tracking for the long term? Which individual is responsible for which action? All of this is being tracked by our memory systems that our agents contribute to.”

Arca 使用 Anthropic 和 OpenAI 等前沿实验室的模型，其中 Claude 是其主要模型提供方。公司打造了自己的模型运行框架，能够根据内部基准测试，自动为特定任务选择最合适的模型。Arca uses models from frontier labs like Anthropic and OpenAI, with Claude its primary model provider. The company has built its own harness, which automatically selects the best model to use for a particular task, based on the company’s internal benchmarks.

拥有自己的运行框架，而不是使用托管式智能体运行框架，也意味着 Arca 可以在需要时灵活更换模型提供商。Owning the harness, rather than using a managed agent harness, also means that Arca has the flexibility to change model providers if it needs to.

对于金融服务而言，安全显然至关重要。Singh 指出，尽管智能体承担了大量工作，但其职责仅限于分析和提出解决方案。交易及任何其他资金转移均由人工顾问处理。Safety is obviously key for financial services. Singh noted that although agents do much of the work, an agent’s role is limited to analysis and proposing solutions. Trading and any other money movement is handled by a human advisor.

我们的所有安全限制都是完全确定性的，”他说。“它们也让智能体无法自行做决定。“All of our safety restrictions are fully deterministic,” he said. “And they take the decision making off the agent's hands.”

换句话说，代理可以自行执行某些工具和任务，但不能执行那些会给客户带来风险的操作。Arca 还会针对已知的 AI 安全问题（例如提示注入）进行验证检查。In other words, there are certain tools and tasks the agent can execute by itself — but not ones that involve risk to the client. Arca also runs verification checks on known AI security issues, such as prompt injections.

Arca 的代理本身并未运行在沙盒中，但会使用沙盒进行计算。Singh 解释说，他们还采取了更进一步的措施，为处理敏感数据增设了额外的安全防护。Arca’s agent isn’t itself sandboxed, but it does use sandboxes for computation. Singh explained they’ve gone even further, with extra security measures for handling sensitive data.

我们构建了一套系统，确保沙箱中的代码能够安全地访问我们的数据，”他说。“我们有这么一个机制，会把它（智能体）编写的代码转换成我们预期的严格格式，然后丢弃原始代码，并在这种格式到达后端之前为其设置防护措施。因此，我们允许智能体使用代码查询我们的数据，这意味着它们可以对数据进行汇总、处理，并按自己的需要加以运用，从而生成诸如财务计划之类的内容。“We've built systems to make it safe for code inside of that sandbox to access our data,” he said. “So we have this thing which translates the code it [the agent] wrote into a strict format that we expect, throw away the code, and you install guardrails on that format before it reaches your back end. So we allow the agents to use code to query our data, which means that they can aggregate it, they can munge it, they can work with it as they see fit to produce [things] like a financial plan.”

Arca 实施的另一个有趣流程是利用人类顾问的反馈以及智能体的执行轨迹来改进 harness。一个“元智能体”会分析所有这些数据，并提出对 harness 的修改建议。Another interesting process Arca has implemented is using feedback from human advisors, along with traces from the agents, to improve the harness. A “meta-agent” is used to analyze all this data and suggest changes to the harness.

播客时间戳Podcast timestamps

• 00:00 — 介绍 Arca 的 AI 原生财富管理公司

  00:00 — Introducing Arca’s AI-native wealth management firm

• 01:40 — AI 如何支持人类财务顾问

  01:40 — How AI supports human financial advisors

• 02:40 — 使用内部基准测试选择模型

  02:40 — Choosing models with internal benchmarks

• 04:01 — 构建自定义测试框架并路由任务

  04:01 — Building a custom harness and routing tasks

• 05:25 — 应对 100,000 美元的税单：智能体与人工审批流程

  05:25 — Handling a $100,000 tax bill: Agents and human approvals

• 07:20 — 长期运行任务中的智能体记忆

  07:20 — Agent memory across long-running tasks

• 09:22 — 安全防护措施与提示注入防御

  09:22 — Safety guardrails and prompt injection defenses

• 11:02 — 默认具备合规性和可审计性

  11:02 — Compliance and auditability by default

• 12:14 —“智能体提示人的次数多于人提示智能体的次数”

  12:14 — “Agents prompt people more than people prompt agents”

• 13:46 — 为什么 Arca 拥有自己的智能体基础设施

  13:46 — Why Arca owns its agent infrastructure

• 15:05 — AI 原生企业与使用 AI 工具的企业

  15:05 — AI-native firms versus firms using AI tools

• 16:09 — 沙箱与财务计划验证

  16:09 — Sandboxes and financial plan verification

• 18:40 — 决策模型与财务事实核查

  18:40 — Decision models and checking financial facts

• 19:51 — 为什么交易执行仍由人类专家负责

  19:51 — Why trade execution stays with human experts

• 20:32 — 将员工反馈转化为评测与改进

  20:32 — Turning staff feedback into evals and improvements

• 22:22 — 将运营知识转化为智能体上下文

  22:22 — Turning operational knowledge into agent context

• 23:41 — 客户案例：隐性费用与主动服务

  23:41 — Client examples: Hidden fees and proactive service

观看阿曼的 AIE 演示文稿Watch Aman’s AIE presentation