随著人工智慧大幅降低软体开发门槛，苹果 App Store 上的应用程式数量虽大幅增加，但在矽谷顶尖新创圈中，原生行动应用程式却迅速边缘化。以指标性育成机构 Y Combinator 最近一期春季班为例，在 195 家入选团队中仅有 8 家将原生行动 App 列为主打产品，比例从十年前的 15% 骤降至不到 1%，显示新创资金与焦点已全面自传统 App 与 SaaS 转向 AI 及代理工作流程。

新一代新创不再仅仅扮演简单转发指令的 GPT 包装器，而是致力于建构能约束、调度并引导多个 AI 模型的控制架构，例如防止 AI 代理误退款或提供虚拟支付凭证等。同时，科技巨头如 Meta 亦推出整合多工任务的智慧代理，推动使用者从在无数零碎 App 间切换，转变为依赖单一中心化系统完成订票、采购或研究等复杂流程。

在商业模式层面，新创公司正从传统订阅费或人头授权费，转向按任务计费的用量模式，透过批发购买底层模型 Token 并交付高质量成果来获取利润。然而，这也意味著软体开发者过去对苹果与 Google 应用程式商店的依赖，如今转移至 OpenAI、Anthropic 等模型供应商手中，技术看门人的集中化挑战依旧存在。

While artificial intelligence has dramatically lowered software development barriers and driven a surge in Apple App Store releases, native mobile applications are rapidly vanishing among Silicon Valley's top startups. In Y Combinator's latest spring cohort of 195 startups, only eight featured a native mobile app as their core offering—a proportion that collapsed from roughly 15% a decade ago to under 1%—signaling a decisive shift in venture focus from apps and SaaS toward AI and agent workflows.

Instead of functioning as trivial GPT wrappers that merely pass text, the new crop of startups is building harness systems that orchestrate, safeguard, and guide AI models across tasks, such as preventing unauthorized refunds or securing financial transactions. Concurrently, tech giants like Meta are introducing all-in-one personal agents, steering users away from hopping across dozens of fragmented apps toward unified systems that execute end-to-end tasks like booking, purchasing, and research.

On the monetization front, startups are departing from conventional app subscriptions or per-seat SaaS licensing in favor of usage-based models, acting as intermediaries that purchase AI tokens wholesale and bill corporate clients per completed task. However, this transition swaps reliance on Apple and Google app stores for dependence on foundational AI providers like OpenAI and Anthropic, shifting the gatekeeping power rather than eliminating it.