首页人物观点我们正在走向一个世界：与软件协作的 AI 智能体数量，将比人类多出 1000 倍。在这样的世界里，“系统记录（systems of record）”的价值不是下降，而是上升。对 AI 智能体的访问权限进行控制、治理其工作流、将其产出整合进更大的业务流程、确保一致性与可靠性等能力，将比“只有人类参与这些流程时”更加重要。只要智能体仍然是现有用户在其工作中的延伸（例如：代表某位用户工作的法律合同智能体），智能体就会与 SaaS 高度互补、并行运行。因此，将会出现三种结果： 1️⃣ 现有系统记录将成为智能体的自然起点在许多既有工作流类别中，只要这些系统能够足够快地移动和适应，它们将成为智能体的发射平台。当智能体要发挥作用，必须依赖大量既有数据和流程时，这种情况尤为明显。其中的关键变量在于：既有厂商能否足够快地转向并演进其平台。 2️⃣ 某些领域将由 AI 原生平台胜出在部分类别中，系统记录要么转型不够快，要么智能体所需的解决方案与原有系统差异过大，AI 原生平台将更好地服务客户。这种情况在以下场景更为明显：缺乏数据或工作流护城河智能体并非现有用户席位的自然延伸我们已经在客户服务以及少数其他领域看到了这一趋势。 3️⃣ 新进入者最大的机会：全新的智能体类别然而，对新公司而言，最大的机会在于那些今天还没有传统软件对应物的全新智能体类别。由于 AI 智能体会把自动化带入几乎所有知识型工作领域，存在大量工作类别并没有天然的软件厂商。这些将主要是：传统上由服务完成的工作或针对“从未有过工具的岗位”的新型软件形态总体结论 AI 智能体 + 软件将协同工作，而这一切在很大程度上只是意味着：整个软件行业的可服务市场（TAM）将迎来一次巨大的扩张。 We’re moving to a world where there will be 1,000x more agents than people working with software. In that world the value of systems of record goes up not down. The ability to control what AI agents have access to, govern their workflows, integrate their work into a broader process, ensure consistency and reliability, and so on, will be even more important than it was when we just had people involved in these workflows. And as long as agents are extensions of existing users in the work they’re doing (a legal contract agent working on behalf of a user), agents will run alongside SaaS in a heavily complementary way. As a result there are 3 outcomes that will happen: 1. Existing systems of record will be the natural launch off point for agents that deal with many existing categories of workflows, assuming they can move and adapt fast enough. This will be the case the more existing data and workflows are required for an agent to be effective; with the X factor being how quickly the incumbent can pivot and evolve their platform. 2. In some categories the system of record won’t move fast enough, or needed solution for agents is so different from the existing system, that AI-native platforms will better serve customers. This will be more pronounced in areas where there’s no data or workflow moat, or where agents aren’t natural extensions of the existing user seats. We’re seeing this in customer service, and a handful of other categories. 3. The biggest play for the new entrants, however, will be all the new categories for AI agents that don’t have a traditional software complement today. Because AI agents will bring automation to all areas of knowledge work, there are many categories of work that don’t have a natural software player. This will mostly be services categories traditionally or new forms of software going after roles that never had tools. Overall, AI agents + software are going to work together and this mostly just represents a major TAM expansion for all software.

我们正在走向一个世界：与软件协作的 AI 智能体数量，将比人类多出 1000 倍。在这样的世界里，“系统记录（systems of record）”的价值不是下降，而是上升。对 AI 智能体的访问权限进行控制、治理其工作流、将其产出整合进更大的业务流程、确保一致性与可靠性等能力，将比“只有人类参与这些流程时”更加重要。只要智能体仍然是现有用户在其工作中的延伸（例如：代表某位用户工作的法律合同智能体），智能体就会与 SaaS 高度互补、并行运行。因此，将会出现三种结果： 1️⃣ 现有系统记录将成为智能体的自然起点在许多既有工作流类别中，只要这些系统能够足够快地移动和适应，它们将成为智能体的发射平台。当智能体要发挥作用，必须依赖大量既有数据和流程时，这种情况尤为明显。其中的关键变量在于：既有厂商能否足够快地转向并演进其平台。 2️⃣ 某些领域将由 AI 原生平台胜出在部分类别中，系统记录要么转型不够快，要么智能体所需的解决方案与原有系统差异过大，AI 原生平台将更好地服务客户。这种情况在以下场景更为明显：缺乏数据或工作流护城河智能体并非现有用户席位的自然延伸我们已经在客户服务以及少数其他领域看到了这一趋势。 3️⃣ 新进入者最大的机会：全新的智能体类别然而，对新公司而言，最大的机会在于那些今天还没有传统软件对应物的全新智能体类别。由于 AI 智能体会把自动化带入几乎所有知识型工作领域，存在大量工作类别并没有天然的软件厂商。这些将主要是：传统上由服务完成的工作或针对“从未有过工具的岗位”的新型软件形态总体结论 AI 智能体 + 软件将协同工作，而这一切在很大程度上只是意味着：整个软件行业的可服务市场（TAM）将迎来一次巨大的扩张。 We’re moving to a world where there will be 1,000x more agents than people working with software. In that world the value of systems of record goes up not down. The ability to control what AI agents have access to, govern their workflows, integrate their work into a broader process, ensure consistency and reliability, and so on, will be even more important than it was when we just had people involved in these workflows. And as long as agents are extensions of existing users in the work they’re doing (a legal contract agent working on behalf of a user), agents will run alongside SaaS in a heavily complementary way. As a result there are 3 outcomes that will happen: 1. Existing systems of record will be the natural launch off point for agents that deal with many existing categories of workflows, assuming they can move and adapt fast enough. This will be the case the more existing data and workflows are required for an agent to be effective; with the X factor being how quickly the incumbent can pivot and evolve their platform. 2. In some categories the system of record won’t move fast enough, or needed solution for agents is so different from the existing system, that AI-native platforms will better serve customers. This will be more pronounced in areas where there’s no data or workflow moat, or where agents aren’t natural extensions of the existing user seats. We’re seeing this in customer service, and a handful of other categories. 3. The biggest play for the new entrants, however, will be all the new categories for AI agents that don’t have a traditional software complement today. Because AI agents will bring automation to all areas of knowledge work, there are many categories of work that don’t have a natural software player. This will mostly be services categories traditionally or new forms of software going after roles that never had tools. Overall, AI agents + software are going to work together and this mostly just represents a major TAM expansion for all software.

Aaron LevieCo-founder & CEO of Box

2025/12/16 22:35 UTC

观点原文

我们正在走向一个世界：

与软件协作的 AI 智能体数量，将比人类多出 1000 倍。

在这样的世界里，“系统记录（systems of record）”的价值不是下降，而是上升。

对 AI 智能体的访问权限进行控制、治理其工作流、将其产出整合进更大的业务流程、确保一致性与可靠性等能力，将比“只有人类参与这些流程时”更加重要。

只要智能体仍然是现有用户在其工作中的延伸（例如：代表某位用户工作的法律合同智能体），智能体就会与 SaaS 高度互补、并行运行。

因此，将会出现三种结果：

1️⃣ 现有系统记录将成为智能体的自然起点

在许多既有工作流类别中，只要这些系统能够足够快地移动和适应，它们将成为智能体的发射平台。

当智能体要发挥作用，必须依赖大量既有数据和流程时，这种情况尤为明显。

其中的关键变量在于：既有厂商能否足够快地转向并演进其平台。

2️⃣ 某些领域将由 AI 原生平台胜出

在部分类别中，系统记录要么转型不够快，要么智能体所需的解决方案与原有系统差异过大，AI 原生平台将更好地服务客户。

这种情况在以下场景更为明显：

缺乏数据或工作流护城河

智能体并非现有用户席位的自然延伸

我们已经在客户服务以及少数其他领域看到了这一趋势。

3️⃣ 新进入者最大的机会：全新的智能体类别

然而，对新公司而言，最大的机会在于那些今天还没有传统软件对应物的全新智能体类别。

由于 AI 智能体会把自动化带入几乎所有知识型工作领域，存在大量工作类别并没有天然的软件厂商。

这些将主要是：

传统上由服务完成的工作

或针对“从未有过工具的岗位”的新型软件形态

总体结论

AI 智能体 + 软件将协同工作，而这一切在很大程度上只是意味着：

整个软件行业的可服务市场（TAM）将迎来一次巨大的扩张。

We’re moving to a world where there will be 1,000x more agents than people working with software.

In that world the value of systems of record goes up not down. The ability to control what AI agents have access to, govern their workflows, integrate their work into a broader process, ensure consistency and reliability, and so on, will be even more important than it was when we just had people involved in these workflows.

And as long as agents are extensions of existing users in the work they’re doing (a legal contract agent working on behalf of a user), agents will run alongside SaaS in a heavily complementary way.

As a result there are 3 outcomes that will happen:

Existing systems of record will be the natural launch off point for agents that deal with many existing categories of workflows, assuming they can move and adapt fast enough. This will be the case the more existing data and workflows are required for an agent to be effective; with the X factor being how quickly the incumbent can pivot and evolve their platform.

In some categories the system of record won’t move fast enough, or needed solution for agents is so different from the existing system, that AI-native platforms will better serve customers. This will be more pronounced in areas where there’s no data or workflow moat, or where agents aren’t natural extensions of the existing user seats. We’re seeing this in customer service, and a handful of other categories.

The biggest play for the new entrants, however, will be all the new categories for AI agents that don’t have a traditional software complement today. Because AI agents will bring automation to all areas of knowledge work, there are many categories of work that don’t have a natural software player. This will mostly be services categories traditionally or new forms of software going after roles that never had tools.

Overall, AI agents + software are going to work together and this mostly just represents a major TAM expansion for all software.

观点原文

相关观点