企业从人工智能实验转向自主运营时，该注意什么？

企业从人工智能实验转向自主运营时，该注意什么？

在过去三年里，企业人工智能对话的焦点是模型可以说什么。 今后三年将讨论允许何种制度。 这更难解决

由Hewlett Packard Entertainment公司赞助，议程的重点是那些将部署和与大赦国际生活在一起的人。 这包括来自梅赛德斯-奔驰北美和其他地方的首席情报官；Booking控股和嘉年华公司的首席警卫官；Estée Lauder Cos.的数据、分析学和AI的主管；以及HPE全球网络化的外地首席技术官James Robertson。

将企业纳入其中很重要。 销售商可以谈论自主操作，但由运营商决定何时以及如何打开。

事件发生在HPE的网络投资人日(Networking Investory Day)一个月后，称 Rami Rahim是HPE网络公司执行副总裁、总裁和总经理。 投资者听到的是一个增长的故事。 CIOs应该把它当作一个操作模型故事来听，因为运行AI的基础设施越来越多地被要求自己运行。

我将关注五件事

问题已经从洞察力转变为行动

generative AI已经做了洞察力表利害关系。 现在大多数企业的建议都比它们能够采取的行动多。 真正的问题不再是“大赦国际能否产生洞察力？” 而是“我能否相信一个大赦国际系统在未经人类同意的情况下，在每一个步骤下采取行动？”

一个起草不良邮件的AI模型创造了一个尴尬的时刻。 AI agent改变网络政策，改变交通路线或对设备进行隔离并获得错误的，可以让企业下线。 我将期待了解各组织如何在它们完全自动化的行动、需要人类批准的行动和仍然手工的行动之间划清界限。

以我的经验来看，最先进的公司并不将自主经营视为即时开关。 他们把它当作一个系统成长成的事物，一种动作。

时间错记分卡

在网络投资者日，HPE的执行副总裁兼校园和分公司总经理Sujai Hajela 提炼了公司自驾车网络背后的理念：“上台不是好的。 ” 任何AI操作讨论都要记住这一点。 一个网络完全可以使用，但仍提供低劣的视频通话，或拖延AI的工作量。

这种从衡量基础设施健康到衡量用户和商业结果的转变是自主运作的先决条件。 如果一个系统无法从用户的角度判断出一个问题是什么样子，你就不能让它自己解决问题。

投资者日最佳例子来自客户。 ServiceNow公司数字核心服务高级主管Sajeev Nair描述了多年的点产品，这些产品一直“在你的遥测中制造另一个已经支离破碎的碎片”。 合并后，他称，人工工作量一度每年耗时2000至3000小时，现在耗时不到60小时。

也希望与业者讨论如何定义代理时代的成功。 我期待答案集中在推动企业前进的结果上，例如回到工程师手中的工时、从未到过票排队的事件

网络正在成为AI的基础设施，而不仅仅是连接

Rahim在投资者日指出，每个为AI融资的CIO都应该认真对待：公司可以在图形处理单位上花费数十亿美元，但如果网络无法满足业绩要求，“这些资源没有得到充分利用 ”。

AE Natarajan经营HPE的路由业务，上下游交通更加平衡，AI交通不能像视频一样被缓存，agents则24小时生成流量。 称表示：「数码用户不会睡着了

当agents成为网络上的新用户，这使得网络成为AI的身份和政策层，而不仅仅是运输层。 基础设施、安全和治理预算也开始趋同。

安全领导人将确定自治的步伐

演讲者排队很有趣：在一次联网活动中，三位演讲者领导安全组织。 在大多数组织中，首席信息安全官员，而不是首席信息干事，将决定如何迅速开展自主行动。

安全领导人面临两面问题。 AI agents需要访问系统、数据，另一方面，攻击者现在也有AI。 兹Q 1Z的Claude神话帮助被审查的组织查明了10 000多个高度或严重的脆弱性，预计这种能力会扩大。 这使得旧的，无支撑的网络齿轮完全在横发。

在我与安保和网络专业人员的谈话中，“Mythos时刻”是环境现代化的警钟。 网络基础设施数量庞大，已有十多年的历史。 这些较老的开关和路由器没有用来防范AI产生的威胁的功能集。

HPE的高级副总裁兼安全和安全总经理David Hughes在投资者日指出，“网络作为感应器”可以探测到威胁，网络本身可以遏制这些威胁，对agents和装置没有信任。 我要知道的是，国际情报和安全局认为自主是管理风险，还是他们需要的辩护。

我的看法是，人类高速安全行动将无法跟上机器高速攻击

阻塞器大多是人和操作的

问从业者什么是减速的代理AI，你很少听到“模型 ”。 你们听到零散的数据、遗留的基础设施、不明确的问责制、技能差距、成本，尤其是信任。

在投资日，HPE将对自驾网络的信任描述为基于证据，系统向运营商展示其建议以及为何在自行行动之前。 这是正确的方法，但它需要干净的遥测技术，许多组织缺乏这种技术。 Rahim还警告说，当买主听到销售商自驾声称时，“你真的必须看盖子下看到底有多少。

我也会在活动上探讨象征漫画的热门话题。 Bloomberg Intelligence软件分析师Niraj Patel在程序上，投资回报对话是许多代理AI项目停顿的地方。 领导人需要表明自主削减成本或解锁新收入，而不仅仅是在试点中发挥作用。

这对企业领导意味着什么？

我建议采取四个步骤：

- 确定你的自主级别。 决定大赦国际可以自行采取哪些行动，哪些行动需要人类批准，哪些行动必须保持人工。 每季度审查清单，并根据结果更新清单。

- 先把数据修好自主操作依赖于一致的，端到端的遥测。 正如ServiceNow所做的那样，整合工具和控制飞机往往是确保一致性的最快方法。

- 给AI agents提供零信任。 对agent行动实行身份意识获取，最低特权政策，持续监测，在网络上实施，而不仅仅是在个人应用中实施。

- 放弃你无法弥补的东西。 清点外援基础设施，并在AI驱动的攻击利用之前更换。

最后想法

向自主操作的移动不会被最佳模式或GPU赢得。 这些企业将建立信任、数据和基础设施，使大赦国际能够安全地采取行动。 因此，带来公司与同一阶段的供应商共同负责营业时间、安全和业务结果。

网络投资者日是一个出色的AI“模拟”日，彭博会活动是主要课程，应当揭示首席信息干事和CISO是否同意。

宙斯·凯拉瓦拉是Kerravala咨询公司下属的一个分部ZK研究公司的主要分析师。 他为SiliconANGLE写了这篇文章。

图片：SiliconANGLE/Gemini

SiliconANGLE联合创始人John Furrier的留言：

支持我们的使命，通过与CUBE社区接触，保持内容的开放和自由。 加入CUBE的校友信托网络，其中技术领导者相互连接，分享情报，创造机会。

- 15M+观看CUBE视频，为AI、云层、网络安全等各种对话提供动力

- 11.4k+CUBE校友——通过独特的信任网络与11400多名技术和企业领袖建立联系，塑造未来

你是AWS客户吗？在财务上支持SiliconANGLE，从我们的市场门户页面和链接购买您的AWS服务：https://siliconangle.com/aws-marketplace/.

关于 SiliconANGLE 媒体

SiliconANGLE Media由技术视觉家约翰·弗瑞尔(John Furrier)和戴夫·韦兰特(Dave Vellante)创立，构建了产业领先数字媒体品牌的动态生态系统，达到1500+百万精英技术专业人士。 利用CUBEAi.com神经网络，帮助科技公司以数据为导向做出决策

本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

What to watch for as enterprises move from AI experimentation to autonomous operations

For the past three years, the enterprise artificial intelligence conversation has focused on what models can say. The next three years will be about what systems are allowed to do. That’s a harder problem that better models alone won’t solve.

On Wednesday, Oct. 28, I’ll be in Boston for the Bloomberg Tech Event: Building the Agentic Enterprise, sponsored by Hewlett Packard Enterprise Co. The agenda focuses on those who will be deploying and living with agentic AI. This includes chief information officers from Mercedes-Benz North America and elsewhere; chief security officers at Booking Holdings and Carnival Corp.; Estée Lauder Cos.’ head of data, analytics and AI; and James Robertson, HPE’s global field chief technology officer for networking.

The inclusion of enterprises is important. Vendors can talk about autonomous operations, but it’s the operators who decide when and how to turn it on.

The event comes a month after HPE’s Networking Investor Day, where the company argued that the network has shifted from plumbing to strategic infrastructure for AI. “AI is reshaping the technology stack, making the network more strategic,” said Rami Rahim, executive vice president, president and general manager of HPE Networking. Investors heard that as a growth story. CIOs should hear it as an operating model story, because the infrastructure that runs AI is increasingly being asked to run itself.

Here are five things I’ll be watching that should matter to CIOs, CTOs and enterprise leadership.

The question has changed from insight to action

Generative AI has made insights table stakes. Most enterprises now have more recommendations than they can act on. The real question is no longer “Can AI generate insights?” but “Can I trust an AI system to take action without human approval at every step?”

An AI model that drafts a bad email creates an awkward moment. An AI agent that changes a network policy, reroutes traffic or quarantines a device and gets it wrong can take a business offline. I’ll be looking to understand how organizations draw the line between actions they fully automate, those that require human approval, and those that remain manual.

In my experience, the most advanced companies don’t treat autonomous operations as an on-off switch. They treat it as something a system grows into, one category of action at a time.

Uptime is the wrong scorecard

At Networking Investor Day, Sujai Hajela, HPE’s executive vice president and general manager of campus and branch, distilled the philosophy behind the company’s self-driving network: “Up is not the same as good.” Keep this in mind for any AI operations discussion. A network can be fully available yet still deliver a poor video call or stall an AI workload.

That shift, from measuring infrastructure health to measuring user and business outcomes, is a prerequisite for autonomous operations. You can’t let a system fix problems on its own if it can’t tell what a problem looks like from the user’s point of view.

The best example at the investor day came from a customer. Sajeev Nair, senior director of digital core services at ServiceNow Inc., described years of point products that kept “creating another fragmentation in your telemetry that is already fragmented.” After consolidating, he said, a manual workload that once took 2,000 to 3,000 hours a year now takes fewer than 60.

At the event, I’m also hoping to talk to practitioners about how they define success in the agentic era. I’m expecting the answers to focus on outcomes that move the business forward, such as hours returned to engineers, incidents that never reach a ticket queue, and faster time to value for new services over headcount reduction.

The network is becoming AI infrastructure, not just connectivity

Rahim made a point at the investor day that every CIO funding AI should take seriously: companies can spend billions on graphics processing units, but if the network can’t meet performance demands, “those resources are not fully utilized.”

AE Natarajan, who runs HPE’s routing business, explained why AI traffic is different. Upstream and downstream traffic become more balanced, AI traffic can’t be cached like video, and agents generate traffic around the clock. “Digital users don’t fall asleep,” he said.

When agents become a new class of user on the network, the network has to understand who they are, what they’re allowed to reach and how they behave. That makes the network an identity and policy layer for AI, not just a transport layer. That’s where infrastructure, security and governance budgets start to converge.

Security leaders will set the pace of autonomy

The speaker lineup is interesting: at a networking event, three speakers lead security organizations. In most organizations, the chief information security officer, rather than the CIO, will decide how quickly autonomous operations proceed.

Security leaders face a two-sided problem. On one side, AI agents need access to systems, data, and each other, creating a growing population of nonhuman identities that traditional zero-trust models weren’t designed for. On the other, attackers now have AI, too. Anthropic’s Claude Mythos has helped vetted organizations identify more than 10,000 high- or critical-severity vulnerabilities, and that capability is expected to spread. That puts old, unsupported network gear squarely in the crosshairs.

In my conversations with security and network professionals, the “Mythos moment” was a wake-up call to modernize the environment. There is a tremendous amount of network infrastructure that is more than a decade old. These older switches and routers do not have the feature set to protect against AI-generated threats.

David Hughes, HPE’s senior vice president and general manager of SASE and security, made the case at investor day that “the network as a sensor” can detect threats and that the network itself can contain them, extending zero trust to agents and devices. I want to hear whether CISOs see autonomy as a risk to manage or a defense they need.

My view is that it’s both. Human-speed security operations won’t keep up with machine-speed attacks, so the same trust model enterprises build for autonomous operations will also determine how well they defend themselves.

The blockers are mostly human and operational

Ask practitioners what’s slowing agentic AI, and you rarely hear “the model.” You hear about fragmented data, legacy infrastructure, unclear accountability, skills gaps, cost and, above all, trust.

At the investor day, HPE described trust in the self-driving network as evidence-based, with the system showing operators what it recommends and why before acting on its own. That’s the right approach, but it requires clean telemetry, which many organizations lack. Rahim also warned that when buyers hear self-driving claims from vendors, “you really have to look under the covers to see how much is actually there.”

At the event, I’ll also be exploring the hot topic of tokenomics. Bloomberg Intelligence software analyst Niraj Patel is on the program, and the return-on-investment conversation is where many agentic AI projects stall. Leaders need to show that autonomy cuts costs or unlocks new revenue, not just that it works in a pilot.

What this means for enterprise leaders

For CIOs and CTOs looking to move from AI pilots to autonomous operations, I’d recommend four steps:

- Define your autonomy tiers. Decide which actions AI can take on its own, which require human approval and which must remain manual. Review the list quarterly and update it based on results.

- Fix the data first. Autonomous operations depend on consistent, end-to-end telemetry. Consolidating tools and control planes, as ServiceNow did, is often the fastest way to ensure that consistency.

- Extend zero trust to AI agents. Apply identity-aware access, least-privilege policies, and continuous monitoring to agent actions, enforcing them across the network rather than only within individual applications.

- Retire what you can’t patch. Inventory out-of-support infrastructure and replace it before AI-driven attacks exploit it.

Final thoughts

The move to autonomous operations won’t be won by the best model or the most GPUs. It will be won by the enterprises that build the trust, data and infrastructure to enable AI to act safely. That’s why this lineup interests me: It brings together the people accountable for uptime, security and business results alongside the vendors on the same stage.

Networking Investor Day was an excellent AI “amuse-bouche” that showed HPE’s self-driving network has moved beyond the slide deck stage. The Bloomberg event is the main course and should reveal whether CIOs and CISO agree.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for SiliconANGLE.

Image: SiliconANGLE/Gemini

A message from John Furrier, co-founder of SiliconANGLE:

Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.

- 15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more

- 11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network

Are you an AWS customer? Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/

About SiliconANGLE Media

Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.

Auto-fetched and translated by AI Atlas. All rights belong to the original author.