夜雨聆风Vibe check from inside one of AI industry's main events: 'Claude mania'
AI行业核心会议现场观察：“Claude狂热”
Apr 11 2026
Key Points 关键要点
本周，超过6500名科技行业高管、创业者与投资人齐聚旧金山HumanX大会，围绕人工智能的最新趋势展开讨论。
Anthropic旗下广受欢迎的编程工具及其强劲的市场势头，成为全场讨论的焦点。
中国在开放权重模型领域的领先优势，也在展会现场被频繁提及，成为重要议题。 |
If one thing became clear at the HumanX conference in San Francisco this week, where 6,500 executives, founders and investors gathered to talk about artificial intelligence, it's that OpenAI no longer dominates the conversation in their industry. For now, at least, that distinction belongs to Anthropic.
本周，6500名高管、创始人与投资人汇聚旧金山HumanX大会，共同探讨人工智能议题。会上有一点愈发明朗：OpenAI已不再主导业界话语权。至少，目前这一位置正被Anthropic取代。
Anthropic's viral coding agent, Claude Code, was the tool on everyone's lips, even as many attendees acknowledged that OpenAI, Cursor and Google are offering strong alternatives.
Claude Code，Anthropic旗下的病毒式传播的编程智能体，成为现场最受关注的工具。尽管与会者普遍认为，OpenAI、Cursor以及谷歌同样提供了具有竞争力的替代方案。
Despite its spat with the Pentagon that went public last month and quickly made its way to the courtroom, Anthropic has only gained momentum. The Department of Defense blacklisted Claude, but after opposing rulings in two courts, Anthropic can keep working with other federal agencies while the cases play out.
尽管Anthropic上个月与美国国防部的争议公开化并迅速进入司法程序，其整体发展势头并未受到明显影响。美国国防部曾将Claude列入限制名单，但在两起法院作出相互不一致的裁决后，Anthropic得以在案件审理期间继续与其他联邦机构开展合作。
Anthropic's early strength in the enterprise has positioned it to benefit from the soaring popularity of AI coding agents, which are used to generate, edit and review code. So while OpenAI kicked off the generative AI boom with the launch of ChatGPT in 2022, Anthropic may be best set up to win contracts from the biggest spenders.
Anthropic在企业市场早期建立的优势，使其得以充分受益于AI编程智能体的快速普及。这类工具被用于代码生成、编辑与审查。因此，尽管OpenAI在2022年凭借ChatGPT引爆生成式AI浪潮，但在争夺大型企业客户合同方面，Anthropic如今可能更具结构性优势。
CNBC spoke with 19 executives and investors at HumanX, some of whom asked not to be named in order to speak freely. Here are the top three takeaways.
在HumanX大会期间，CNBC采访了19位企业高管与投资人，其中部分人士因希望畅所欲言而要求匿名。以下为采访的三大核心要点。
Claude has 'become a religion'
Claude已经变成一种信仰
Anthropic was founded in 2021 by a group of researchers and executives who defected from OpenAI. The startup is valued at $380 billion, making it one of the most valuable private companies in the world.
Anthropic由一批从OpenAI出走的研究人员和高管于2021年创立，目前估值达3800亿美元，跻身全球最具价值的私营企业之列。
Claude Code launched to the general public in May 2025, and as of February was generating more than $2.5 billion in annualized revenue. Arvind Jain, CEO of enterprise AI company Glean, said Claude Code has inspired "Claude Mania," which is putting pressure on business leaders to deploy it.
Claude Code于2025年5月向公众发布。截至今年2月，其年化收入已超过25亿美元。企业AI公司Glean首席执行官阿尔文德·贾恩（Arvind Jain）表示，Claude Code正在催生一种“Claude狂热”，并正在对企业决策者形成部署压力。
"It has become a religion, that's the level of that mania," Jain said in an interview. "Everybody, if you go and ask them today, 'Hey, if I gave you one AI tool, what tool would you want?' The answer would be Claude."
“它已经到了某种宗教级别的热度。”贾恩在采访中表示：“你现在去问任何人：‘如果只能选一个AI工具，你会选哪个？’答案几乎都是Claude。”
On Tuesday, Anthropic announced a new AI model, Claude Mythos Preview, with advanced cybersecurity capabilities thanks to its strong coding and reasoning skills. The model sparked a lot of buzz at HumanX, even though its rollout is limited to a select group of roughly 50 companies.
周二，Anthropic发布了新模型Claude Mythos Preview。该模型因其强大的编程与推理能力，进一步强化了其在网络安全领域的应用潜力。目前该版本仅向约50家精选企业开放，但在HumanX大会上引发了广泛关注。
Victor Riparbelli, CEO of AI video company Synthesia, said Anthropic has managed to demonstrate focus and restraint with its models and product, which can be difficult for a young hyper-growth company.
AI视频公司Synthesia首席执行官维克多·里帕贝利（Victor Riparbelli）表示，Anthropic在模型与产品策略上展现出明显的专注与克制，而这对于一家高速成长的年轻科技公司而言并不容易。
"The guys at Anthropic were just like, 'We're not going to do anything about video, we're not going to care about voice models, we're just going to solve code gen,' and now we're here," Riparbelli said in an interview. "OpenAI has had the problem of having to market six different products, which just takes up mind space for the consumer."
“Anthropic团队的策略很明确：不做视频，不做语音模型，只专注解决代码生成问题。结果你现在看到的就是他们的成果。”里帕贝利表示：“相比之下，OpenAI不得不同时推广六款不同产品，这会分散用户注意力，占据大量心智空间。”
One investor cautioned that while Anthropic has been consistent and managed to identify a sticky AI use case, the industry is still young, and momentum could easily swing in another direction.
一位投资人则提醒称，尽管Anthropic在方向上保持稳定，并成功抓住了一个具有高黏性的AI应用场景，但整个行业仍处于早期阶段，市场动能随时可能发生转移。
AI change management
AI变革管理
As tech companies work to usher their customers into the AI era, they're also grappling with how to leverage and deploy agents internally. Even for Silicon Valley startups, keeping up with the pace of change is no easy feat.
随着科技公司推动客户全面进入AI时代，它们自身也在摸索如何在内部有效利用并部署AI智能体。即便是硅谷初创企业，也在努力应对这一快速演进的技术节奏。
Ashwin Sreenivas, president of AI startup Decagon, said the advent of coding agents has led to a number of shifts within his company. Decagon has changed its interview process to allow candidates to use the tools, and the company is able to rely on smaller teams of engineers.
AI初创公司Decagon总裁阿什温·斯里尼瓦斯（Ashwin Sreenivas）表示，编程智能体的出现正在深刻改变公司内部运作方式。Decagon已调整招聘流程，允许候选人在面试中使用这些工具，同时公司也因此能够依赖更小规模的工程师团队。
A project that may have required four or five engineers "becomes two engineers because everyone can move a lot faster and go a lot farther," Sreenivas said in an interview.
“原本需要四五名工程师才能完成的项目，现在两个人就能搞定，因为每个人的效率和能力边界都被显著拉高。”斯里尼瓦斯在采访中表示。
For Navrina Singh, CEO of AI governance startup Credo AI, the proliferation of new AI tools has been simultaneously exciting and anxiety inducing for her. Overcommunicating, particularly with her customers, has become essential, she said.
对于AI治理初创公司Credo AI的CEO 纳夫里娜·辛格（Navrina Singh）而言，新AI工具的快速涌现既令人兴奋，也带来持续的焦虑。她表示，加强沟通，尤其是与客户之间的沟通，已成为企业运营的关键环节。
"The things that I could not do last year and I needed to hire 10 people, I can actually build over a weekend and deploy for myself and for the company," Singh said. "The anxiety is I can't control my roadmap, and I can't control my commitments to the enterprise customers who love more clarity and who like a little bit more stability."
辛格表示：“过去一年做不到、需要雇佣10个人才能完成的事情，现在我一个周末就能独立搭建并部署到公司内部使用。但令我焦虑的是，我无法完全掌控产品路线图，也难以稳定兑现对企业客户的承诺，而这些客户更偏好清晰度与稳定性。”
Big tech incumbents are navigating similar changes.
大型科技巨头同样在应对类似的变革。
Cisco President Jeetu Patel said roughly 85% of his company's engineering workforce, or about 18,000 employees, are using AI, but the path to getting there was unlike what he'd anticipated. Patel said Cisco initially learned it needed to prioritize adoption over outcomes, and to trust that model capabilities will continue to improve.
思科总裁吉图·帕特尔（Jeetu Patel）表示，公司约85%的工程师团队（约1.8万人）已经在使用AI工具，但这一转型路径与他最初的预期截然不同。帕特尔指出，思科最初意识到必须优先推动“采用率”，而非单纯关注短期产出，同时也需要相信模型能力仍将持续提升。
"You can't think of these as tools, you have to think of these as digital coworkers that are joining your team, because your composition of your scrum team changes," Patel said at the conference. "You might not have a scrum team of eight people. You might have a scrum team of two people and six agents, or two people and infinite agents."
“你不能再把这些当作工具来看待，而应将其视为加入团队的数字同事，因为你的敏捷开发团队结构已经发生变化。”帕特尔在大会上表示：“团队可能不再是八个人，而是两个人加六个智能体，甚至是两个人加无限数量的智能体。”
The race
与中国AI模型竞速
The fragile two-week ceasefire agreement between the U.S. and Iran has massive implications for energy and financial markets across the globe. But the vast majority of execs and investors who spoke to CNBC at HumanX this week said they're not yet experiencing any direct business impact from the latest conflict in the Middle East.
美伊之间脆弱的两周停火协议对全球能源与金融市场具有深远影响。然而，本周在HumanX大会接受CNBC采访的绝大多数企业高管与投资人表示，他们尚未感受到中东最新冲突对自身业务的直接影响。
Rather, they're focused on another looming geopolitical problem: China's open-weight models.
相较之下，他们更关注另一个正在浮现的地缘技术竞争焦点：来自中国的开放权重模型。
In AI, a model is considered open weight if its parameters, or the elements that improve its outputs and predictions during training, are publicly available. As of April, Chinese open-weight models, including GLM-5.1, Kimi K2.5 and Qwen3.5, dominate industry benchmarks.
在AI领域，如果一个模型的参数（即在训练过程中用于优化输出与预测能力的核心权重）可以被公开获取，则被称为“开放权重（open-weight）”模型。截至4月，中国的多个开放权重模型，包括GLM-5.1、Kimi K2.5以及Qwen3.5，已在多项行业基准测试中占据领先位置。
American companies are swarming to China's models. Cursor built its Composer 2 model using Kimi 2.5. Airbnb CEO Brian Chesky told CNBC in October that his company's chatbot was largely dependent on Alibaba's Qwen.
与此同时，美国企业正加速采用中国模型。Cursor基于Kimi 2.5构建了其Composer 2模型；Airbnb首席执行官布莱恩·切斯基（Brian Chesky）曾在去年10月向CNBC表示，公司聊天机器人在很大程度上依赖阿里巴巴的Qwen模型。
Given the importance the U.S. AI industry is placing on beating China when it comes to innovation, there's a big emphasis domestically on closing the gap in open weight. Two investors told CNBC they're dedicating a lot of their time and resources to that effort, and a third said it's one of the key problems for the industry to solve right now.
鉴于美国AI行业高度重视在技术创新层面超越中国模型，国内正大力强调缩小开放权重领域的差距。两位投资人告诉CNBC，他们正投入大量时间与资源于此；第三位投资人则表示，这是当下行业亟需破解的核心难题之一。
Glean's Jain said having multiple options is critical.
Glean公司的贾恩表示，拥有多元选择至关重要。
"The trend that we see is that enterprises today, they're very wary of depending on one or two providers for all of their AI," Jain said. "They don't want to work with just one model company, because they know that innovation is happening across many and also in open source. You want to have a choice."
贾恩表示：“我们观察到的趋势是，企业越来越不愿意将全部AI能力押注在一两家供应商身上。他们不想只依赖单一模型公司，因为创新正在多个公司以及开源生态中同时发生。企业需要保有选择权。”
