具体数据上，上周五深度心智推出「精灵三号」世界构建工具后，游戏相关股票受挫，其中泰克图互动在消息冲击下下跌8%。更剧烈的波动出现在周二：安思罗匹克为其克劳德协作平台发布「法律」外挂后，市场在当日收盘前引发跨软体、金融服务与资产管理等板块合计约2850亿美元的市值蒸发；周三又出现第二波抛售，一档偏重软体的交易所交易基金下跌1.8%，并录得连续第7个交易日下跌。摩根大通分析师托比·奥格以「未审先判」形容情绪，指出即使业绩超预期也不足以扭转定价；杰富瑞交易员杰弗里·法武扎则以「下一个印刷媒体或百货业」作类比，强调悲观钟摆已大幅摆向全面抛售。
作者反驳市场把短期演示等同于可销售、可托付的企业级产品，强调从「有用」到「可信」仍需大量验证，尤其牵涉法律责任时更不可能在「一个周末加几个工作小时」内完成替代决策；安思罗匹克自身仍使用沃克戴等企业订阅也被用来佐证替换门槛。摩根士丹利将新能力视为加剧竞争、对汤森路透等专业公司偏负面；同时作者以自动驾驶为对照，指出距离完全自主的最后几层改进呈指数级变难。泰克图互动执行长斯特劳斯·泽尔尼克认为人工智慧可提升效率但无法捕捉剧情、情感连结、氛围与任务结构等核心体验；结论是：法律外挂可能像「顶尖助理」般改善一致性与格式，但在反复警示勿盲信的前提下，市场反应把「可用」直接等同「可信」的幅度过大。
The article argues that AI has become an “existential threat” to the software industry, and that the market’s reaction has quickly leapt from wowed-by-demos to a broad valuation reset. Breakthroughs in chat models followed by rapid progress in image and video generation are treated as evidence that software-tool boundaries are being steadily eroded. Investors therefore price the uncertainty with a “sell first, ask later” mindset, effectively discounting near-term unknowns straight into stock prices.
In concrete terms, last Friday Google DeepMind unveiled the “Genie 3” world-building tool, hitting game-related stocks; Take-Two Interactive fell 8% on the news. The sharper shock came on Tuesday after Anthropic released a “Legal” plug-in for its Claude Cowork collaboration platform: by the close, selling spread across software, financial services, and asset management, totaling about $285 billion in market value erased. A second wave followed on Wednesday, when a software-heavy ETF dropped 1.8% and logged its 7th straight down day; JPMorgan analyst Toby Ogg described sentiment as being “sentenced before trial,” while Jefferies trader Jeffrey Favuzza likened the mood to software becoming the next print media or department stores as pessimism swung toward indiscriminate selling.
The author pushes back against equating short-term demonstrations with sellable, dependable enterprise products, stressing that the gap from “useful” to “trustworthy” requires extensive validation and is even less likely to be closed “in one weekend and a handful of working hours” when legal liability is involved; Anthropic itself still uses Workday and other enterprise subscriptions as a signal that switching barriers remain high. Morgan Stanley frames the new capabilities as intensifying competition and potentially negative for specialists like Thomson Reuters, while the author compares the situation to self-driving cars, where the last layers of improvement become exponentially harder. Take-Two CEO Strauss Zelnick says AI can improve efficiency but cannot capture core experiences like storyline, emotional connection, vibe, and mission structure; likewise, the Legal plug-in may function like a top-tier assistant that improves consistency and formatting, but given repeated warnings not to blindly trust outputs, the market’s leap from “usable” to “reliable” is portrayed as far too large.