专家向 Axios 表示，AI 模型开发商之间的竞赛将在 2026 年持续，同时，AI 智能体数量将进一步增加，企业也将面临越来越大的压力，需要证明 AI 能在现实世界中带来实际回报。
Why it matters: AI may be both the current and next big thing, but success increasingly hinges less on being the "best" model and more on timing.为何重要： AI 也许既是当下的风口，也是下一个风口，但成功越来越不取决于是否拥有“最强模型”，而更多取决于是否把握好时机。
The big picture: Rapid progress by OpenAI, Anthropic, Google and others drove frequent leapfrogging — and fierce price competition — in 2025. That dynamic is expected to intensify next year and beyond.宏观来看： OpenAI、Anthropic、谷歌等公司在 2025 年的快速进展推动了频繁的技术超越以及激烈的价格竞争，这种态势预计将在明年及以后进一步加剧。
"We're just gonna be in this constant race," Box CEO Aaron Levie told Axios.Box 首席执行官亚伦·莱维（Aaron Levie）对 Axios 表示：“我们将一直处在这种持续不断的竞赛之中。”
Reality check: There are important, often-overlooked steps between the arrival of more powerful algorithms and a boost in productivity.现实检验： 在更强大的算法出现与生产力真正提升之间，存在许多重要却常被忽视的环节。
The winners must understand when a technology is mature enough to deploy and how to integrate it into messy, human-run organizations without burning money or credibility.真正的赢家必须明白技术何时成熟到可以部署，以及如何在不浪费资金、不损害信誉的前提下，将其融入由人类主导、运作复杂的组织中。
"Good AI won't need long prompts. The more you have to explain, the worse the product is." Winston Weinberg, CEO and co-founder of Harvey, tells Axios. "The best systems will already know the context."Harvey 首席执行官兼联合创始人温斯顿·温伯格（Winston Weinberg）对 Axios 表示：“好的 AI 不需要很长的提示词。你需要解释得越多，产品就越糟。最好的系统应该已经理解上下文。”
"A jump in model capability does not instantly mean that task gets automated in the economy," Levie said. "There's still a lot of work and software that has to get built out from there."莱维表示：“模型能力的提升并不意味着相关任务会立刻在经济中实现自动化。从那一步到真正落地之间，还有大量工作和软件需要构建。”
Case in point: Coding has been among the earliest and biggest beneficiaries of generative AI for a simple reason: the work is already structured for it. It's largely text-based, modular, and designed around tight human-machine feedback loops.典型例子： 编程是生成式 AI 最早、也是受益最大的领域之一，原因很简单——这类工作本身就高度结构化。它主要基于文本、模块化，并围绕紧密的人机反馈循环而设计。
"You have the perfect workflow in coding," Levie says. And then you have knowledge work, "which is 10 times messier than what engineering workflows look like."莱维说：“编程拥有几乎完美的工作流程。”而相比之下，知识型工作“比工程工作流程要混乱十倍”。
Between the lines: Semi-autonomous agents were the talk of 2025, but businesses were hesitant to hand off too much work to AI models that were still prone to making mistakes.言外之意： 半自主智能体是 2025 年的热门话题，但由于 AI 模型仍然容易出错，企业对将过多工作交给它们仍持谨慎态度。
Improved models could help make agents more of a reality.模型能力的提升，可能会让智能体真正变得可行。
"A year from now, answering questions will be the least useful thing AI can do. (And it will be excellent at answering questions!)," Fidji Simo, OpenAI CEO of applications, tells Axios.OpenAI 应用业务首席执行官菲吉·西莫（Fidji Simo）对 Axios 表示：“一年之后，回答问题将成为 AI 最不重要的能力之一——尽管它在回答问题方面会非常出色。”
"Instead, we'll have proactive AI assistants constantly running in the background, getting things done for us across the web and the real world," she says. "It will anticipate our needs, and we'll be able to trust it to make decisions and take action on our behalf."她说：“取而代之的将是主动型 AI 助手，它们会在后台持续运行，在网络和现实世界中替我们完成任务。它们会预判我们的需求，我们也能够信任它们代我们做决定并采取行动。”
Zoom in: Willem Avé, head of product at Square, agrees that agents will continue to be more trustworthy and capable. And "companies implementing AI will get more creative about connecting them to deterministic systems that will take the variability out of AI results," Avé tells Axios.进一步来看： Square 产品负责人威廉·阿维（Willem Avé）也认为，智能体将变得越来越可靠、能力越来越强。他对 Axios 表示，部署 AI 的公司将更加富有创造性地把智能体与确定性系统连接起来，从而减少 AI 输出结果的不稳定性。
"In 2026, the most successful companies will set goals that sound absurd without AI—and then use agent collaboration to make them routine," Dan Rogers, CEO of Asana, tells Axios.Asana 首席执行官丹·罗杰斯（Dan Rogers）对 Axios 表示：“到 2026 年，最成功的公司将制定一些如果没有 AI 听起来近乎荒谬的目标，然后通过智能体协作把这些目标变成日常操作。”
"The litmus test is simple: if your 2026 targets could have been your 2024 targets, you're not thinking ambitiously enough about what agents working in concert can actually unlock," per Rogers.罗杰斯表示：“判断标准很简单：如果你的 2026 年目标在 2024 年就已经可以提出，那说明你还没有充分想象多智能体协同工作真正能够释放的潜力。”
Yes, but: Businesses could be in for another year of messy agent rollouts.但另一方面： 企业在智能体部署方面，可能还将经历一年混乱的推进过程。
Ryan Gavin, CMO of Slack at Salesforce, predicts that "2026 will be the year of the lonely agent."Salesforce 旗下 Slack 的首席营销官瑞安·加文（Ryan Gavin）预测：“2026 年将是‘孤独智能体之年’。”
Gavin says companies will spin out "hundreds of agents per employee," but most will sit idle, like unused software licenses: "impressive but invisible," Gavin says.加文表示，企业可能会为每位员工部署“数百个智能体”，但其中大多数将像未使用的软件许可证一样闲置——“看起来很厉害，却几乎不可见”。
"In an agentic solution, you're breaking down the problem into many, many steps. And the overall solution is only accurate if you're accurate each step of the way," AT&T chief data officer Andy Markus told Axios. "That's the challenge."AT&T 首席数据官安迪·马库斯（Andy Markus）对 Axios 表示：“在智能体解决方案中，你需要把问题拆解成非常多的步骤，而只有每一步都准确，整体结果才会准确。这正是挑战所在。”
The need for financial payoff was a consistent theme among the experts we asked to predict AI trends for 2026.在我们采访的所有预测 2026 年 AI 趋势的专家中，“财务回报”的需求是一个反复出现的主题。
"2026 is the 'show me the money' year for AI," Venky Ganesan, a partner at Menlo Ventures, tells Axios. "Enterprises will need to see real ROI in their spend, and countries need to see meaningful increases in productivity growth to keep the AI spend and infrastructure going."Menlo Ventures 合伙人文基·加内森（Venky Ganesan）对 Axios 表示：“2026 年将是 AI 的‘拿出真金白银’之年。企业必须看到真实的投资回报率，国家层面也必须看到生产率的显著提升，AI 投入和基础设施建设才能持续下去。”
2026 will also be the year when pragmatism supplants optimism, James Brundage, leader of EY's Global and Americas Technology Sector, tells Axios. "Boards will stop counting tokens and pilots and start counting dollars."安永全球及美洲科技行业负责人詹姆斯·布伦戴奇（James Brundage）对 Axios 表示，2026 年也将是务实取代乐观的一年，“董事会将不再统计 token 数量和试点项目，而是开始计算真金白银的回报”。
Ganesan predicts that some of the aggressive spending could bankrupt major companies.加内森预测，一些激进的支出策略可能会导致大型公司破产。
The other side: "We will also see a major AI company have an IPO and GDP growth numbers will go up in America by over 100 basis points," Ganesan says.另一面： 加内森表示：“我们也将看到一家大型 AI 公司进行 IPO，美国的 GDP 增长率将提高超过 100 个基点。”
The bottom line: The pace of AI adoption in business continues to be limited by the ability of humans (and organizations made up of humans) to adapt.总结： AI 在企业中的应用速度，仍然受限于人类自身（以及由人类组成的组织）的适应能力。
"The agents that matter will show up where work happens, understand context, and just work," says Gavin.加文表示：“真正重要的智能体会出现在工作的发生地，理解上下文，并且能真正发挥作用。”
