内容来源：https://www.technologyreview.com/2026/03/04/1133642/bridging-the-operational-ai-gap/

内容总结：

企业AI部署面临“运营鸿沟”，一体化集成平台成破局关键

尽管人工智能的变革潜力已成共识，众多企业也已从试点转向生产部署，但全面实现运营成功仍面临挑战。近期由MIT Technology Review Insights进行的一项调查揭示了关键症结：缺乏稳固的运营基础。

调查访问了500名美国中大型企业的IT高管，发现虽然76%的受访公司至少有一个部门已部署AI生产流程，但企业级广泛落地仍存障碍。高达40%的受访企业表示，其AI项目因成本、准确性与治理问题难以推进。研究指出，核心问题并非AI技术本身，而在于缺失的数据、系统集成与自动化工作流基础。

报告关键发现包括：

1. 成功有径可循：近半数（43%）的成功AI应用基于定义清晰、已自动化的流程。
2. 团队建设滞后：仅三分之一（34%）的企业设有专职团队维护AI工作流。
3. 集成平台价值凸显：采用企业级一体化集成平台的公司，在AI工作流中使用5种以上数据源的可能性高出5倍。这些企业也更易实现跨部门AI部署、赋予工作流更高自主性，并对未来扩大应用更具信心。

专家指出，随着智能体AI的兴起，模型自主性不断增强，对数据、应用和系统进行一体化集成的全局观变得前所未有的重要。缺乏这一基础，企业AI项目可能停滞于试点或面临失败风险。构建坚实的集成运营底座，已成为企业跨越AI“运营鸿沟”、释放智能体工作流潜力的先决条件。

中文翻译：

赞助内容
弥合运营性人工智能的鸿沟
企业正借助全域集成，将当前的流程自动化延伸至未来的智能体工作流。

与Celigo合作呈现

人工智能的变革潜力已得到广泛认可。企业应用案例正蓄势待发，各组织正从试点项目转向生产环境中的人工智能。企业不再仅仅谈论人工智能，而是正在调整预算和资源以付诸实践。许多企业已在尝试智能体人工智能，它有望实现新水平的自动化。然而，对许多企业而言，实现全面运营成功的道路仍不明朗。尽管人工智能试验遍地开花，但企业范围内的全面采用依然难以实现。

如果没有集成的数据和系统、稳定的自动化工作流以及治理模型，人工智能项目就可能困在试点阶段，难以投入生产。智能体人工智能的兴起和模型自主性的日益增强，使得采取整体性方法集成数据、应用和系统变得比以往任何时候都更加重要。缺乏这一点，企业人工智能计划就可能失败。Gartner预测，到2027年，超过40%的智能体人工智能项目将因成本、准确性和治理挑战而被取消。真正的问题并非人工智能本身，而是缺失的运营基础。

为了解各组织如何构建其人工智能运营体系以及如何部署成功的人工智能项目，《麻省理工科技评论》洞察团队于2025年12月，对美国500名大中型企业的高级IT领导者进行了调查，所有这些企业都以某种方式在推进人工智能。

调查结果以及一系列专家访谈（均于2025年12月进行）表明，强大的集成基础与更先进的人工智能实施相一致，有助于企业范围内的计划推进。随着人工智能技术和应用的发展与普及，集成平台可以帮助组织避免重复建设和数据孤岛，并在应对工作流日益增长的自主性时保持清晰的监管。

报告的主要发现包括：

部分组织在人工智能方面取得进展。近年来，多项研究揭示了人工智能缺乏实质性成功。然而，我们的研究发现，四分之三（76%）的受访公司至少有一个部门拥有人工智能工作流完全投入生产。

人工智能在定义明确、既有的流程中最易成功。近半数（43%）的组织在将人工智能应用于定义明确且自动化的流程时取得成功。四分之一的组织在新流程上取得成功。另有三分之一（32%）的组织正将人工智能应用于各种流程。

三分之二的组织缺乏专门的人工智能团队。只有三分之一（34%）的组织拥有专门维护人工智能工作流的团队。五分之一（21%）的组织表示中央IT部门负责持续的人工智能维护，25%的组织表示责任在于部门运营。对于19%的组织，责任是分散的。

企业级集成平台带来更稳健的人工智能实施。拥有企业级集成平台的公司，在其人工智能工作流中使用更多样化数据源的可能性高出五倍。十分之六（59%）的公司使用了五个或更多数据源，相比之下，仅针对特定工作流使用集成的组织中这一比例仅为11%，而未使用集成平台的组织则为0%。使用集成平台的组织还拥有更多跨部门的人工智能实施、人工智能工作流中更高的自主性，以及对未来赋予自主性更强的信心。

本内容由《麻省理工科技评论》旗下定制内容部门Insights制作。并非由《麻省理工科技评论》的编辑人员撰写。内容由人类作者、编辑、分析师和插画师进行调研、设计和撰写。这包括调查问卷的编写和调查数据的收集。可能使用的人工智能工具仅限于经过严格人工审查的次要生产流程。

深度报道
人工智能
一场“QuitGPT”运动正敦促人们取消ChatGPT订阅
对ICE的强烈反对正在推动一场更广泛的运动，反对人工智能公司与特朗普总统的联系。
Moltbook是人工智能戏剧的巅峰时刻
这个病毒式传播的机器人社交网络，在揭示智能体未来的同时，更多地反映了我们自身当前对人工智能的狂热。
认识将大语言模型当作外星生物研究的新一代生物学家
通过将大语言模型当作活物而非计算机程序来研究，科学家们首次发现了它们的一些秘密。
2026年人工智能的下一步是什么
我们的AI作者对来年做出了大胆预测——以下是值得关注的五大热门趋势。

保持联系
获取《麻省理工科技评论》的最新动态
发现特别优惠、头条新闻、即将举办的活动等更多内容。

英文来源：

Sponsored
Bridging the operational AI gap
Enterprise-wide integration is being leveraged to extend today’s process automations into tomorrow’s agentic workflows.
In partnership withCeligo
The transformational potential of AI is already well established. Enterprise use cases are building momentum and organizations are transitioning from pilot projects to AI in production. Companies are no longer just talking about AI; they are redirecting budgets and resources to make it happen. Many are already experimenting with agentic AI, which promises new levels of automation. Yet, the road to full operational success is still uncertain for many. And, while AI experimentation is everywhere, enterprise-wide adoption remains elusive.
Without integrated data and systems, stable automated workflows, and governance models, AI initiatives can get stuck in pilots and struggle to move into production. The rise of agentic AI and increasing model autonomy make a holistic approach to integrating data, applications, and systems more important than ever. Without it, enterprise AI initiatives may fail. Gartner predicts over 40% of agentic AI projects will be cancelled by 2027 due to cost, inaccuracy, and governance challenges. The real issue is not the AI itself, but the missing operational foundation.
To understand how organizations are structuring their AI operations and how they are deploying successful AI projects, MIT Technology Review Insights surveyed 500 senior IT leaders at mid- to large-size companies in the US, all of which are pursuing AI in some way.
The results of the survey, along with a series of expert interviews, all conducted in December 2025, show that a strong integration foundation aligns with more advanced AI implementations, conducive to enterprise-wide initiatives. As AI technologies and applications evolve and proliferate, an integration platform can help organizations avoid duplication and silos, and have clear oversight as they navigate the growing autonomy of workflows.
Key findings from the report include the following:
Some organizations are making progress with AI. In recent years, study after study has exposed a lack of tangible AI success. Yet, our research finds three in four (76%) surveyed companies have at least one department with an AI workflow fully in production.
AI succeeds most frequently with well-defined, established processes. Nearly half (43%) of organizations are finding success with AI implementations applied to well-defined and automated processes. A quarter are succeeding with new processes. And one-third (32%) are applying AI to various processes.
Two-thirds of organizations lack dedicated AI teams. Only one in three (34%) organizations have a team specifically for maintaining AI workflows. One in five (21%) say central IT is responsible for ongoing AI maintenance, and 25% say the responsibility lies with departmental operations. For 19% of organizations, the responsibility is spread out.
Enterprise-wide integration platforms lead to more robust implementation of AI. Companies with enterprise-wide integration platforms are five times more likely to use more diverse data sources in AI workflows. Six in 10 (59%) employ five or more data sources, compared to only 11% of organizations using integration for specific workflows, or 0% of those not using an integration platform. Organizations using integration platforms also have more multi-departmental implementation of AI, more autonomy in AI workflows, and more confidence in assigning autonomy in the future.
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 “QuitGPT” campaign is urging people to cancel their ChatGPT subscriptions
Backlash against ICE is fueling a broader movement against AI companies’ ties to President Trump.
Moltbook was peak AI theater
The viral social network for bots reveals more about our own current mania for AI as it does about the future of agents.
Meet the new biologists treating LLMs like aliens
By studying large language models as if they were living things instead of computer programs, scientists are discovering some of their secrets for the first time.
What’s next for AI in 2026
Our AI writers make their big bets for the coming year—here are five hot trends to watch.
Stay connected
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.