内容来源：https://www.technologyreview.com/2026/08/12/1141032/scaling-ai-agents-with-trustworthy-data/

内容总结：

AI智能体规模化应用的关键：突破遗留数据系统瓶颈
——调查报告显示，数据领先型企业正通过重塑数据基础释放智能体潜力

随着生成式AI向自主决策与行动加速演进，企业数据系统正面临前所未有的压力。一项针对300名数据与科技高管的全球调查显示，当前AI智能体平均仅能访问企业45%的数据，而“数据落后型”企业中这一比例不足30%。然而，少数“数据领先型”企业已将数据接入率提升至70%以上，并在智能体应用中取得显著成效。

遗留系统成主要障碍
调查指出，传统遗留数据系统——即使是数年前更新过的系统——难以满足智能体对实时、跨部门、多模态数据的需求。三分之二的数据落后型企业表示，遗留系统限制了智能体的扩展能力（66%）并阻碍其快速决策（68%）。相比之下，数据领先型企业中仅有8%报告存在此类问题。

信任源自数据基础
目前，仅约半数受访企业信任其AI智能体决策的准确性与相关性，而数据领先型企业中这一比例达到100%。报告强调，可靠的AI必须建立在可靠的数据基础之上，数据就绪度直接决定智能体决策的可信度。

紧迫的时间窗口
调查显示，所有受访企业均计划在未来两年内部署智能体AI，其中69%预计将广泛使用。若无法消除数据系统瓶颈，智能体AI将难以兑现其承诺的速度与效率。数据接入与上下文理解被列为规模化应用的首要任务，其次为增强数据与AI治理中的业务语境，以及推动数据管理自动化。

领先者的启示
报告认为，数据领先型企业为行业提供了可借鉴的路径：他们不仅优先打通结构化与非结构化数据的访问通道，还通过自动化数据管理和业务上下文注入，构建了智能体“茁壮成长”的生态环境。这些企业的实践证明，唯有彻底摆脱遗留系统的束缚，才能实现智能体AI的规模化部署与可信运转。

（本文基于MIT Technology Review Insights部门与Google Cloud合作发布的调查报告，内容经人工研究与撰写，AI工具仅用于辅助制作流程。）

中文翻译：

赞助内容

用可信数据扩展AI智能体规模

企业如何摆脱传统数据系统的束缚，为AI智能体提供动力，实现可信的自主行动。

与谷歌云联合呈现

业务和技术领导者们无需被说服便已确信，智能体AI时代已经到来。各组织正在快速采用智能体，很少有高管质疑这项技术改变工作方式的潜力。但许多组织发现，要从AI中获得理想的投资回报率，关键在于拥有正确的基础设施，而基础设施不足和数据问题正是主要障碍。

智能体AI对企业数据系统提出了相当大的新要求。从回答问题转向采取行动，意味着AI智能体需要来自全企业的数据，涵盖所有结构化和非结构化形式，并具备正确的业务上下文。为了实时做出决策并采取行动，智能体还需要无障碍地访问组织的运营系统——例如存储供应链、销售点或人力资源数据的系统。传统数据系统，即使是几年前刚更新的系统，也难以满足这些要求。

随着AI智能体更广泛地嵌入企业运营，克服传统数据系统限制的需求变得越发紧迫。如果Gartner预测的“到2027年AI智能体将增强或自动化50%的业务决策”被证明是正确的，那么组织必须消除瓶颈，否则就有可能让智能体无法获得快速做出正确决策所需的数据。

本报告基于对300位数据和技术高管的调查，探讨了传统系统如何限制了许多组织中AI智能体的有效性。报告发现，少数组织——即数据领先者——在智能体AI方面取得了更大的成功，并且因传统系统而遭遇的数据限制较少。这些领先者为创造合适的数据环境、让智能体蓬勃发展以及让可信系统实现规模化提供了指南。

报告的主要发现包括：

目前很少有公司为智能体AI提供充足的企业数据访问权限。在所有受访组织中，AI平均只能访问公司数据的45%。在被归类为“数据落后型”的组织中，这一数字降至30%或更低。然而，有一小部分组织确保了对超过70%数据的访问。这些“数据领先者”的智能体表现比其他组织更为成功。

对智能体决策的信任反映了数据就绪程度。如今，只有约一半的受访组织相信其AI智能体做出的决策准确且相关。相比之下，100%的数据领先者信任其智能体的决策，这强有力地表明可靠的AI需要可靠的数据基础。

数据领先者更容易实现智能体的规模和速度。三分之二的数据落后型组织表示，传统数据系统限制了AI智能体的扩展（66%），并阻碍了智能体快速做出决策（68%）。由于基本上已经克服了传统数据的限制，数据领先者大多已扫清这些障碍，只有8%报告存在任一限制。

让数据资产为智能体做好准备的压力迫在眉睫。在两年内，100%的受访者计划使用智能体AI，其中69%预计将广泛使用。如果不消除数据系统的限制，智能体AI将无法实现其承诺的速度和效率。

数据访问和上下文是首要任务。在所有受访者中，实现扩展最重要的举措是改善AI智能体对结构化和非结构化数据的访问。同样位居前列的还有通过业务上下文增强数据和AI治理。数据领先者还高度关注数据管理的自动化。

本内容由MIT Technology Review旗下定制内容部门Insights制作。并非由MIT Technology Review编辑团队撰写。由人类作者、编辑、分析师和插画师完成研究、设计和撰写，包括问卷设计和调查数据收集。可能使用的AI工具仅限于经过全面人工审核的次级制作流程。

深度探索

人工智能

一家初创公司声称突破了制约大语言模型的瓶颈

Subquadratic现已分享了其新模型的更多细节。但仍有一些人持怀疑态度。

一个根本性缺陷使大语言模型极易遭受攻击

这使得诱骗它们做不该做的事情变得很容易，例如告诉您如何破坏飞机的导航系统。

Anthropic发现了一个隐藏空间，Claude在其中推敲概念

一项新技术使该公司能够比以往任何时候都更深入地探究大语言模型的奇异工作机制。

Claude Science是Anthropic最新推出的旗舰产品

该公司正加大对AI用于科学研究的投入。

保持联系

获取来自
MIT Technology Review
的最新资讯

发现特别优惠、热门故事、即将举行的活动等更多内容。

英文来源：

Sponsored
Scaling AI agents with trustworthy data
How companies are freeing themselves of legacy data systems to power AI agents that deliver trusted, autonomous action.
In partnership withGoogle Cloud
Business and technology leaders need no convincing that the time of agentic AI is here. Organizations are rapidly adopting agents, and few executives doubt the technology’s potential to transform work. But many organizations find that realizing the desired return on investment (ROI) from AI hinges on having the right foundation, with inadequate infrastructure and data being major blockers.
Agentic AI places considerable new demands on enterprise data systems. The shift from answering questions to taking actions means AI agents need data from across the enterprise, in all its structured and unstructured forms, and with the right business context. To make decisions and act in real time, agents also need frictionless access to the organization’s operational systems—for example, those storing its supply chain, point-of-sale, or human resources data. Legacy data systems, even those updated just a few years ago, struggle to meet these demands.
As AI agents become embedded more widely in enterprise operations, the need to overcome the restrictions of legacy data systems grows more urgent. If Gartner’s prediction that AI agents will augment or automate 50% of business decisions by 2027 proves correct, organizations must eliminate bottlenecks or risk depriving agents of the data they need to make the right decisions at speed.
This report, based on a survey of 300 data and technology executives, explores how legacy systems are limiting the effectiveness of AI agents in many organizations. It finds that a handful of organizations—the data leaders—are having greater success with agentic AI and experiencing fewer data limitations as a result of legacy systems. These leaders offer a guide to creating the right data environment for agents to flourish and trusted systems to scale.
Key findings from the report include:
Few companies currently provide agentic AI with ample access to enterprise data. Across all the surveyed organizations, AI only has access to an average of 45% of company data. That number falls to 30% or less in organizations categorized as “data laggards”. A select group, however, ensures access to over 70% of their data. These “data leaders” are having greater success with their agents than the rest.
Trust in agent decisions is a reflection of data readiness. Today, only around half of surveyed organizations trust that the decisions their AI agents make are accurate and relevant. By contrast, 100% of the data leaders trust their agents’ decisions, a strong indicator that reliable AI requires a reliable data foundation.
Data leaders find it easier to achieve agent scale and speed. Two-thirds of data laggards say legacy data systems limit AI agent scaling (66%) and prevent agents from making decisions at speed (68%). Having largely overcome legacy data constraints, the leaders have mostly cleared these roadblocks, with just 8% reporting either constraint.
The pressure is on to make data estates agent-ready. Within two years, 100% of respondents plan to be using agentic AI, with 69% expecting to use it widely. Without removing data system constraints, agentic AI will fail to deliver the desired speed and efficiencies it promises.
Data access and context are top priorities. The most important initiative to enable scaling among all respondents is improving access to structured and unstructured data for AI agents. Also high on the list is enhancing data and AI governance with business context. Data leaders are also focusing heavily on the automation of data management.
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
Deep Dive
Artificial intelligence
A startup claims it broke through a bottleneck that’s holding back LLMs
Subquadratic has now shared more details about its new model. But some are still skeptical.
A fundamental flaw leaves LLMs strikingly vulnerable to attack
It makes it easy to trick them into doing things they shouldn’t, such as telling you how to sabotage an aircraft’s navigation system.
Anthropic found a hidden space where Claude puzzles over concepts
A new technique has let the company probe deeper than ever into the weird workings of an LLM.
Claude Science is Anthropic’s newest flagship product
The company is doubling down on AI for science.
Stay connected
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.