以下是这段内容的中文翻译：
Decagon 的创始人 Jesse 和 Ashwin 对不断演进的 AI 格局提出了独特的见解，他们坚称，即便在前沿模型和通用人工智能 (AGI) 兴起之时，Decagon 公司的相关性依然持久。他们详细阐述了 Decagon 从依赖大型闭源模型（OpenAI、Anthropic）转向在其 90% 的运营中利用开源替代方案的历程。这一转变是由降低延迟、增强控制和提高成本效益的关键需求驱动的，尤其对他们的语音代理而言。他们反对常见的“智能与成本”权衡论，证明经过特定任务微调的更小开源模型，在专门化应用时，其性能、速度和成本均可超越大型通用模型。这种专业化方法需要一个专门的研发团队进行持续微调，Decagon 将其视为一项战略投资。前沿模型则保留用于复杂的探索性辅助功能，例如 Decagon 的“Duet Autopilot”，它负责分析对话并提出系统改进建议。
为了反驳“最后的创业公司”这一论调（即前沿实验室将垄断所有 AI 应用），Decagon 坚称应用层具有持久的必要性。即使拥有先进的 AGI，代理也需要结构化环境来存储信息、检索数据和执行复杂推理。应用程序提供了整合业务逻辑、管理工作流程、确保合规性以及提供垂直特定解决方案的必要基础设施——这些是单纯模型智能无法提供的要素。Decagon 将自己定位为“代理实验室”，致力于构建这些全面、企业级的应用堆栈，这些堆栈对于单个公司在内部开发而言过于专业。
他们的“前线部署工程师”（FDEs）发挥着关键作用，他们超越了单纯的咨询，转变为识别客户需求并将解决方案产品化以实现更广泛的适用性。一个例子是“Duet Autopilot”，它自动化了代理操作程序 (AOPs) 的创建、系统集成、测试和对话监控——这些任务以前都是手动执行的。这一策略体现了 Decagon 以产品为导向的理念，旨在通过深入的客户参与来构建可扩展的解决方案。
即使在 AGI 的未来，Decagon 的长期“护城河”也存在于这种强大的软件基础设施中。这包括用于定义代理能力和约束、实现协作开发、确保监管合规性以及从大量对话数据中提取可操作洞察的系统。这种企业就绪的框架是使 AI 能够在复杂组织结构中部署和管理的关键。
在向大型企业销售时，Decagon 倡导“玻璃盒”方法，为客户提供对其 AI 代理的透明度和控制权，这与更不透明的“黑盒”解决方案形成对比。这促进了更快的迭代和更大的客户自主性。创始人 Jesse 和 Ashwin 深度参与销售，加速流程并确保客户的直接反馈能指导产品开发，着重于高影响力的用例以建立发展势头。
Decagon 的产品愿景已从客户支持扩展到更广泛的“AI 礼宾”角色，能够处理如入站销售和运营工作流等多样化的业务流程。这一演进得益于 AI 模型在遵循复杂指令和有效进行开放式对话方面能力的提升。
关于运营挑战，Decagon 认为招聘顶尖人才而非技术限制是其主要瓶颈。他们认为，AI 能力的提升会带来更高的生产力，促使公司进行更多建设，从而招聘更多而非更少的人。在文化方面，他们将苛刻的工作环境描述为共同抱负和团队导向方法的自然结果，优先考虑沟通和协作，而非强制延长工时。这种文化通过结构化入职培训和经验丰富的员工部署到新的国际办事处而实现全球拓展。
最后，关于 AI 对就业的影响，Decagon 认为 AI 自动化了繁琐的工作，“淘汰了职位而非职业”。这使得人类能够从事更有价值、更具创造性和战略性的工作，常常导致“杰文斯悖论”——即服务成本降低会增加需求，从而创造新的机会。例如，CRM 系统并非过时，而是成为 AI 代理交互的关键数据存储库，从而提升了它们的价值。
Decagon's founders, Jesse and Ashwin, offer a unique perspective on the evolving AI landscape, asserting their company's enduring relevance amidst the rise of frontier models and AGI. They detail Decagon's journey from relying on large, closed-source models (OpenAI, Anthropic) to leveraging open-source alternatives for 90% of their operations. This shift was driven by the critical need for lower latency, greater control, and cost-efficiency, especially for their voice agents. They argue against the common "smart vs. cheap" tradeoff, demonstrating that smaller, fine-tuned open-source models, when specialized for specific tasks, can outperform larger, general-purpose models in performance, speed, and cost. This specialized approach necessitates a dedicated research team for continuous fine-tuning, which Decagon views as a strategic investment. Frontier models are reserved for complex, exploratory auxiliary functions, such as Decagon's "Duet Autopilot" which analyzes conversations and suggests system improvements.
Countering the "last startups" narrative that posits frontier labs will monopolize all AI applications, Decagon asserts the enduring necessity of application layers. Even with advanced AGI, agents require structured environments to store information, retrieve data, and execute complex reasoning. Applications provide the essential infrastructure for integrating business logic, managing workflows, ensuring compliance, and offering vertical-specific solutions—elements that raw model intelligence alone cannot deliver. Decagon positions itself as an "Agent Lab" that builds these comprehensive, enterprise-grade application stacks, too specialized for individual companies to develop internally.
Their "forward-deployed engineers" (FDEs) play a pivotal role, evolving beyond mere consulting to identifying customer needs and productizing solutions for broader applicability. An example is "Duet Autopilot," which automates the creation of Agent Operating Procedures (AOPs), system integrations, tests, and conversation monitoring – tasks previously performed manually. This strategy reflects Decagon's product-led philosophy, aiming to build scalable solutions from deep customer engagement.
Decagon's long-term "moat," even in an AGI future, resides in this robust software infrastructure. This includes systems for defining agent capabilities and constraints, enabling collaborative development, ensuring regulatory compliance, and extracting actionable insights from vast conversational data. This enterprise-ready framework is key to making AI deployable and manageable within complex organizational structures.
In selling to large enterprises, Decagon champions a "glass box" approach, offering customers transparency and control over their AI agents, contrasting with more opaque "black box" solutions. This fosters faster iteration and greater customer autonomy. Founders Jesse and Ashwin are deeply involved in sales, accelerating processes and ensuring direct customer feedback informs product development, focusing on high-impact use cases to build momentum.
Decagon's product vision has expanded from customer support to a broader "AI concierge" role, capable of handling diverse business processes like inbound sales and operational workflows. This evolution is powered by AI models' improved ability to follow complex instructions and engage in open-ended conversations effectively.
Regarding operational challenges, Decagon identifies hiring top talent as its primary bottleneck, rather than technological limitations. They contend that AI's increased capabilities lead to higher productivity, prompting companies to build more and, consequently, hire more, not less. On culture, they describe their demanding environment as a natural outcome of shared ambition and a team-oriented approach, prioritizing communication and collaboration over mandated long hours. This culture is scaled globally through structured onboarding and experienced personnel deploying to new international offices.
Finally, addressing AI's impact on employment, Decagon argues that AI automates mundane tasks, "killing jobs but not careers." This frees humans for more valuable, creative, and strategic work, often leading to "Jevin's paradox" where reduced service costs increase demand, creating new opportunities. CRMs, for example, are not obsolete but become crucial data repositories that AI agents interact with, enhancing their value.