The Old Software Moat Is Dead — How to Spot the Enterprise AI Companies That Are Actually Winning

Tungsten Automation 首席人工智能官 Adam Field 认为，仅仅提供强大的 AI 工具是不够的；真正的竞争优势源于有效的实施和深厚的领域专业知识。Tungsten Automation 为包括 40% 财富 100 强企业在内的超过 25,000 家组织提供服务，并观察了企业在数字化转型中成败的原因。 Field 强调，行业正经历从“老派”的确定性自动化（例如机器人流程自动化，RPA，即系统重复预定义任务）到如今 AI 驱动的工作流引擎的根本性转变。现代“智能体”可以被赋予一个目标，并利用可用的工具和信息来实现它，这代表着能力的巨大飞跃。 尽管市场关注芯片制造商（英伟达、AMD）和大型语言模型（如 Anthropic、OpenAI 等 LLM），Field 强调，对企业而言，真正的价值在于其上构建的**自动化层**。对于许多日常任务，例如处理发票，所使用的特定 LLM 将成为一种商品。竞争优势源于围绕这些模型构建的行业基础、专有数据和专业应用。 AI 本身并不新鲜；Tungsten 已经使用它 40 年了。Field 倡导“用正确的 AI 完成正确的工作”，强调传统的机器学习 (ML) 技术仍然具有巨大的价值。它们通常比大型生成式 AI 模型更快、更具成本效益且更环保。他认为，最佳方法是一个智能结合 ML 和生成式 AI（GenAI）来解决特定问题的混合平台。 Field 提出的一个关键概念是“暗数据”——即组织中 80% 或更多的非结构化信息，这些信息被困在文档、电子邮件、通话记录和合同中。历史上，这些数据对机器或人类而言都无法进行分析。AI 现在使企业能够阅读和处理这些复杂的非结构化文档，并将这些信息与结构化数据相结合，从而在客户服务、投资和风险评估等领域做出更好的决策。这种能力直接影响企业的增长轨迹和效率，通过减少商业贷款申请等流程中的摩擦，缩短了现金周转周期。 然而，在试点阶段之后成功实施 AI 是一项挑战。许多企业 AI 项目失败的原因是，试点项目通常侧重于“容易的 80%”——即理想情况——而忽略了“另外的 20%”的复杂性，例如处理异常情况、低质量输入（如倾斜拍摄的身份证照片），或基础模型可能被弃用的动态性质。扩展 AI 需要大量的人力资本、严格的测试和能够管理这些实际复杂性的强大平台。Field 认为，企业应该利用像 Tungsten 这样的专业平台来处理复杂任务，让内部 AI 专家能够专注于业务的差异化。 关于“SaaS 危机”和传统 SaaS 业务的竞争护城河，Field 认为，随着数据迁移变得更容易，作为“记录系统”的旧护城河正在被侵蚀。新的护城河建立在开发解决实际问题的优秀技术、深厚的行业专业知识（例如 Tungsten 40 年的经验），以及至关重要的，将**风险**从客户那里**转移**（例如为发票处理管理 140 个国家的合规性）的基础上。 为了区分价值与炒作，Field 建议超越营销流行语。他创造了“无聊 AI”（"boring AI"）一词，指的是需要时间构建但对于取得有意义成果至关重要的基础性工作（数据基础设施、安全性、合规性）。投资者不应仅仅关注效率或成本降低，而应寻找那些能够展示具体**成果**、提高**熟练度**和**创造收入**的公司。他指出，那些在 AI 方面进行大量投资**同时**增加员工数量的公司，更有可能将 AI 用于增长而非仅仅削减成本，这表明其策略更具可持续性和影响力。 最后，Field 对 AI 的未来既表示谨慎又充满兴奋。他主要关注的是**安全**问题——数据隐私、恶意代码注入的可能性以及对国家安全的更广泛影响。积极的一面是，他从 AI 普及**教育**的潜力中获得了启发，尤其是在服务不足的地区。AI 工具可以扩大教师的覆盖范围，提供以前不可能实现的优质学习机会，并使个人能够更有效地获取知识。

Adam Field, Chief AI Officer at Tungsten Automation, argues that simply providing powerful AI tools isn't enough; true competitive advantage comes from effective implementation and deep domain expertise. Tungsten Automation, serving over 25,000 organizations including 40% of the Fortune 100, has observed how companies succeed or fail in digital transformation. Field highlights a fundamental shift from "old school" deterministic automation (like Robotic Process Automation, RPA, where systems repeat predefined tasks) to today's AI-driven workflow engines. Modern "agents" can be given a goal and use available tools and information to achieve it, representing a significant leap in capability. While the market focuses on chip manufacturers (NVIDIA, AMD) and large language models (LLMs like Anthropic, OpenAI), Field stresses that the real value for enterprises lies in the *automation layer* built on top. For many routine tasks, like processing an invoice, the specific LLM used becomes a commodity. The competitive edge is derived from the industry foundations, proprietary data, and specialized applications built around these models. AI itself isn't new; Tungsten has been using it for 40 years. Field advocates for using the "right AI for the right job," emphasizing that traditional machine learning (ML) technologies still offer tremendous value. They are often faster, more cost-effective, and environmentally friendly than large generative AI models. The optimal approach, he suggests, is a hybrid platform that intelligently combines ML and GenAI to solve specific problems. A crucial concept Field introduces is "dark data"—the 80% or more of an organization's information that is unstructured and trapped in documents, emails, call transcripts, and contracts. Historically, this data was inaccessible to machines or humans for analysis. AI now enables companies to read and process these complex, unstructured documents, integrating this information with structured data to make better decisions in areas like customer service, investment, and risk assessment. This ability directly impacts a company's growth trajectory and efficiency, shrinking cash conversion cycles by reducing friction in processes like commercial loan applications. However, implementing AI successfully beyond pilot phases is a challenge. Many enterprise AI projects fail because pilots often focus on the "easy 80%"—ideal scenarios—while ignoring the complexities of the "other 20%," such as handling exceptions, low-quality inputs (e.g., sideways photos of IDs), or the dynamic nature of foundational models that can be deprecated. Scaling AI requires significant human capital, rigorous testing, and robust platforms that can manage these real-world intricacies. Field argues that companies should leverage specialized platforms like Tungsten's for complex tasks, allowing their internal AI experts to focus on differentiating the business. Regarding the "SaaS-pocalypse" and the competitive moats of traditional SaaS businesses, Field believes the old moat of being a "system of record" is eroding as data migration becomes easier. New moats are built on developing great technology that solves real problems, deep industry know-how (e.g., Tungsten's 40 years of experience), and crucially, *transferring risk* from clients (such as managing compliance across 140 countries for invoice processing). To discern value from hype, Field suggests looking beyond marketing buzzwords. He coined the term "boring AI," referring to the foundational work (data infrastructure, security, compliance) that takes time to build but is essential for meaningful outcomes. Instead of focusing solely on efficiency or cost reduction, investors should look for companies that demonstrate tangible *outcomes*, increased *proficiency*, and *revenue generation*. He notes that companies making significant AI investments *and* increasing headcount are likely using AI for growth, rather than just cost-cutting, indicating a more sustainable and impactful strategy. Finally, Field expresses both caution and excitement about AI's future. His main concern revolves around *security*—data privacy, the potential for nefarious code injection, and the broader implications for national security. On the positive side, he finds inspiration in AI's potential to democratize *education*, particularly in underserved regions. AI tools can multiply the reach of teachers, providing access to quality learning that was previously impossible, and empowering individuals to acquire knowledge more effectively.