AI在编程以外领域推广缓慢的根本原因并非技术本身不足，而是将实验性技术嵌入企业陈旧的业务流程需要耗费大量时间与精力。客户服务领域的AI代理大多仅能进行对话，因为它们未被授予存取企业关键系统的权限，无法真正采取行动。与此同时，云端巨头如Google、Amazon和Microsoft出于留住客户的考量，有时甚至鼓励企业使用更便宜的模型，而非最昂贵的前沿产品，这进一步加剧了AI实验室的商业压力。
面对AI编程市场的价格竞争与营收集中风险，业界寄望于「杰文斯悖论」——当某项技术变得更便宜时，总体使用量反而会因更多人受到经济激励而增长。然而，该悖论并不保证市场扩大后的赢家会是OpenAI或Anthropic等前沿实验室，廉价开源模型同样可能抢占份额。正如互联网泡沫所示，持续的需求增长并不能确保每个参与者都能存活，AI编程虽是当今技术革命的基石，但对某些企业而言，仅靠编程这一支柱或许远远不够。
Three years into the AI boom, corporate use of advanced frontier models remains overwhelmingly concentrated on software coding, which accounts for an estimated one-third to one-half of enterprise token consumption—and by some measures even more. Other applications such as customer service, back-office operations, accounting, and HR occupy much smaller shares of spending, revealing that a supposedly transformative technology is still relatively narrow in its real-world adoption.
The slow spread of AI beyond coding stems not from technological shortcomings but from the difficulty of integrating experimental tools into legacy business workflows. Most customer-service AI agents deployed today function as little more than chatbots because they lack the permissions to act within corporate systems. Meanwhile, cloud giants like Google, Amazon, and Microsoft sometimes steer clients toward cheaper or open-source models to keep them locked into cloud platforms, adding commercial pressure on frontier labs such as OpenAI and Anthropic.
Industry optimists invoke Jevons Paradox—the idea that as AI coding tools become cheaper through competition, overall usage will rise because more people are economically incentivized to adopt them. Yet the paradox does not guarantee which players will capture the growing demand; low-cost open-source alternatives like DeepSeek and Alibaba's Qwen could claim significant market share. As the dotcom bust demonstrated, sustained demand growth does not ensure survival for every ambitious participant, and relying predominantly on coding revenue may prove insufficient for the AI labs seeking to justify their enormous valuations.