OpenAI发布了企业级Agent平台Presence，直接接入企业客户服务、销售、账单处理等核心流程[1]。这一举动标志着Agent技术从生成层走向应用层，传统SaaS领域面临冲击。受此影响，HubSpot股价下跌超过12%，Atlassian跌幅接近12%，其他软件公司股价也出现明显下滑[1]。
Agent应用的爆发式增长正在重塑基础设施需求。OpenAI企业客户API推理Token消耗在一年内增长约320倍，预计2026年4月处理规模将超每分钟150亿Token[1]。这种增长源于Agent任务中Token消耗可能比传统聊天场景高出百倍、千倍[1]，相应推动了芯片等基础设施的需求。SK海力士因此在2026年Q2营收和营业利润创历史新高，主要受HBM和AI服务器DRAM需求推动[1]。
尽管垂直Agent应用面临通用模型能力补齐的压力，但行业专业知识、流程和数据仍具有不可替代性[1]。围绕Agent的新型经济体系正在形成，垂直应用开发商虽然面临短期冲击，但仍保有差异化竞争空间。
OpenAI has unveiled an enterprise-grade Agent platform called Presence, marking a significant shift as Agent technology moves from the model layer into real-world applications [1]. The platform directly integrates into core enterprise functions including customer service, sales, and billing processes [1].
The launch has already rattled the software industry. HubSpot's stock price fell more than 12%, while Atlassian declined nearly 12%, with other software companies experiencing notable share price drops [1]. This market reaction reflects growing concerns among traditional SaaS providers facing competition from generalist AI systems now capable of handling enterprise workflows.
The surge in Agent adoption is reshaping infrastructure demands. OpenAI's enterprise customers have seen their API reasoning token consumption increase approximately 320-fold over one year, with processing volumes projected to exceed 15 billion tokens per minute by April 2026 [1]. Agent tasks can consume tokens at rates 100 to 1,000 times higher than conventional chat scenarios [1], fundamentally altering computational requirements across the industry.
Despite these pressures, vertical Agent applications retain competitive advantages rooted in domain-specific knowledge, workflow expertise, and proprietary data that general-purpose models cannot easily replicate [1]. Industry analysts suggest the market will eventually stabilize around a new economic structure centered on Agent capabilities, where specialized vertical solutions coexist with broader platforms by leveraging their distinct informational and procedural advantages [1].