OpenAI’s Sales Chief Sees ‘Paradigm Shift’ in Corporate AI Spending
OpenAI 的销售负责人看到企业人工智能支出的“范式转变”

To win more such corporate deals, it’s been hiring sales staff. The number has risen to nearly 300 people, up from roughly 200 in June, making up nearly one-fifth of OpenAI’s 1,600-person staff. Lionetti, who joined OpenAI in July, previously was chief revenue officer at Zapier after serving in sales roles at enterprise software firms Confluent, Dropbox and Atlassian.
为了赢得更多此类企业交易，OpenAI 一直在招聘销售人员。人数已增加到近 300 人，较 6 月份的约 200 人有所上升，几乎占 OpenAI 1,600 名员工的五分之一。Lionetti 于 7 月加入 OpenAI，此前曾担任 Zapier 的首席收入官，并在企业软件公司 Confluent、Dropbox 和 Atlassian 担任销售职务。

OpenAI’s goal is to prove so useful to customers that its tools become a regular, recurring part of their budgets. Many companies who started testing its products in 2023 have gradually increased their spending over the past year, but that spending has been primarily limited to specific projects within companies as executives evaluate return on investment, Lionetti said.
OpenAI 的目标是对客户提供如此有用的服务，以至于其工具成为他们预算中常规、重复的一部分。Lionetti 表示，许多在 2023 年开始测试其产品的公司在过去一年中逐渐增加了支出，但这些支出主要限于公司内部的特定项目，因为高管们在评估投资回报。

More recently, customers have shown interest in expanding their use of ChatGPT and other OpenAI products across their entire company, he added. For instance, T-Mobile has signed a deal with OpenAI to use its artificial intelligence across all of the telecom company’s customer-facing products, Lionetti said. That deal is worth $100 million over three years, The Information reported.
最近，客户对在整个公司范围内扩大使用 ChatGPT 和其他 OpenAI 产品表现出了兴趣，他补充道。例如，T-Mobile 与 OpenAI 签署了一项协议，将其人工智能应用于该电信公司所有面向客户的产品，Lionetti 表示。该协议的价值为 1 亿美元，期限为三年，《信息》报道。

OpenAI competes for enterprise contracts against incumbents like Google and Microsoft, it has to counteract their sales pitches, which can highlight their own longstanding security and compliance guarantees.
OpenAI 在与谷歌和微软等现有企业竞争合同时，必须对抗它们的销售宣传，这些宣传可能会强调它们长期以来的安全性和合规性保证。

“If you look at our road map forward, security is a huge focus for us, and that’s how we represent it to our customers as well,” Lionetti said.
“如果你看看我们的未来路线图，安全是我们非常关注的重点，这也是我们向客户展示的方式，”Lionetti 说。

While businesses pay for access to OpenAI’s application programming interface and ChatGPT, OpenAI expects payments for chatbot subscriptions will generate most of its revenue for years to come.
虽然企业为访问 OpenAI 的应用程序编程接口和 ChatGPT 付费，但 OpenAI 预计聊天机器人订阅的付款将在未来几年内产生大部分收入。

It’s also expecting more sales from new products, such as o1, its new reasoning model, which can analyze term sheets and legal contracts, simplify manufacturing processes, and evaluate health claims for insurance companies, among other tasks, Lionetti said.
Lionetti 表示，该公司还预计新产品的销售将增加，例如 o1，这是一种新的推理模型，可以分析条款清单和法律合同，简化制造流程，并评估保险公司的健康声明等任务。

In the interview, Lionetti also spoke about how the startup is thinking about its consumer businesses and pricing its cutting-edge o1 models, its partnership with Apple and its plans for forthcoming capabilities like agents.
在采访中，Lionetti 还谈到了初创公司如何考虑其消费业务以及为其尖端 o1 模型定价，与苹果的合作以及未来功能如代理的计划。

This interview has been edited for length and clarity.
本次采访经过编辑，以便于长度和清晰度。

How are you thinking about OpenAI’s growth strategy now? Are you focusing more on ChatGPT Teams and Enterprise more than the application programming interface, or where are you seeing the most growth among enterprise customers?
您现在对 OpenAI 的增长战略有什么看法？您是否更关注 ChatGPT 团队和企业版，而不是应用程序编程接口，或者您在企业客户中看到的增长主要在哪里？

OpenAI has broad capabilities that are really designed to grow across a number of verticals. ChatGPT is where most of our users will start today in the business world. That will help you with a number of things—think productivity, internal operations—that’s where ChatGPT will be valuable to you as an organization. But then there’s this whole other world of custom development, and that’s where the APIs come in. So it’s not an either/or for our customers, it’s a both, and it’s really just trying to figure out, OK, what is your use case? What are you trying to solve?
OpenAI 拥有广泛的能力，旨在跨多个垂直领域发展。ChatGPT 是我们大多数用户今天在商业世界中的起点。这将帮助您处理许多事务——想想生产力、内部运营——这就是 ChatGPT 作为组织对您有价值的地方。但还有一个完全不同的定制开发世界，这就是 API 的作用。因此，对于我们的客户来说，这不是二选一，而是两者兼具，关键在于弄清楚，好的，您的使用案例是什么？您想解决什么问题？

Do you find that the customers who use the API need to be more technically savvy, or need to have internal data science or engineering teams, than those who would just use an out-of-the-box tool like ChatGPT?
您是否发现使用 API 的客户需要比仅使用像 ChatGPT 这样的现成工具的客户更具技术能力，或者需要拥有内部数据科学或工程团队？

That’s a good question, so when you think about the API business itself, it is really custom. It’s flexible to your needs, and that’s the way we’ve created the APIs. We are starting to make capabilities in our APIs a little easier, so it becomes a little more self-serve, and we give you more flexibility to do that, but at the end of the day we find that customers who are looking at our API do have some of those internal [engineering] teams, and if they don’t we actually partner with companies like a Bain or a PwC to help them get going with their deployments as well.
这是个好问题，所以当你考虑 API 业务本身时，它实际上是定制的。它灵活地满足你的需求，这就是我们创建 API 的方式。我们开始让我们的 API 中的功能变得更简单一些，这样它就变得更自助，我们给你更多的灵活性来做到这一点，但归根结底，我们发现那些关注我们 API 的客户确实有一些内部的[工程]团队，如果没有，我们实际上会与像贝恩或普华永道这样的公司合作，帮助他们启动部署。

But what ChatGPT also offers is a level of flexibility—we have something called GPTs, so think of that as a shell of ChatGPT, and you can give it information that can be translated to your personal use cases. As an example, our finance team internally uses GPTs for some data analysis.
但 ChatGPT 还提供了一种灵活性——我们有一种叫做 GPT 的东西，可以将其视为 ChatGPT 的外壳，您可以提供信息，以便将其转化为您的个人用例。例如，我们的财务团队在内部使用 GPT 进行一些数据分析。

I’ve spoken to some companies who are using ChatGPT as a multipurpose catch-all tool to replace other types of legacy software. Are you seeing that type of traction with large enterprises?
我与一些公司进行了交谈，他们正在使用 ChatGPT 作为一种多功能的通用工具，以取代其他类型的传统软件。您在大型企业中看到这种趋势吗？

We believe AI products are truly a paradigm shift, and that starts to unlock these new ways of working that you’re referring to here. Users and customers are finding a new way to work.
我们相信人工智能产品确实是一个范式转变，这开始解锁您在这里提到的新工作方式。用户和客户正在找到一种新的工作方式。

There are organizations like Moderna using ChatGPT Enterprise and using GPTs across their whole company to analyze clinical data and visualize data sets. You have retail organizations like Lowes using our APIs to help employees identify product description errors and improve the customer search experience. And then [Arizona State University] is using ChatGPT to do teaching, research, operations, and they’re starting to power their educational and learning experience across the board.
有像 Moderna 这样的组织在使用 ChatGPT Enterprise，并在整个公司中使用 GPT 来分析临床数据和可视化数据集。还有像 Lowes 这样的零售组织在使用我们的 API 来帮助员工识别产品描述错误并改善客户搜索体验。然后，[亚利桑那州立大学]正在使用 ChatGPT 进行教学、研究和运营，并开始在各个方面提升他们的教育和学习体验。

So we don’t want to think too far ahead, but when you think of AI as a technology, we’ve been at the frontier of that, and the whole goal is to have a true paradigm shift, and that’s what you’re seeing.
所以我们不想考虑得太远，但当你把人工智能视为一种技术时，我们一直处于这一领域的前沿，整个目标是实现真正的范式转变，这就是你所看到的。

Are those customers you mentioned spending substantially on your products? How are they thinking about return on investment from that spending?
您提到的那些客户在您的产品上花费很大吗？他们如何看待这些支出的投资回报？

We can’t give out information about what they’re spending, but the way we work with customers on ROI is pretty straightforward—we actually want to solve their complex use cases, and some of them are very quantifiable. So there are the T-Mobiles of the world; they’re leveraging it for something called IntentCX to power all their customer experiences through AI, and some of that is very quantifiable in the customer service realm or the go-to-market realm. With the likes of Moderna, if you think about them, it comes down to accuracy and speed to market, and we’re hopefully helping them do that as well.
我们无法提供他们支出的具体信息，但我们与客户在投资回报率（ROI）方面的合作方式相当简单——我们实际上希望解决他们复杂的使用案例，其中一些是非常可量化的。因此，像 T-Mobile 这样的公司正在利用一种叫做 IntentCX 的技术，通过人工智能推动他们的所有客户体验，其中一些在客户服务领域或市场推广领域是非常可量化的。以 Moderna 为例，如果你考虑他们的情况，关键在于准确性和市场速度，我们希望也能帮助他们实现这一点。

I’ve also heard of some companies who are using OpenAI’s models to replicate or replace enterprise software from vendors like Salesforce or Workday. Have you been seeing movement in that direction with customers?
我也听说过一些公司正在使用 OpenAI 的模型来复制或替代像 Salesforce 或 Workday 这样的供应商的企业软件。您在客户方面是否看到朝这个方向的动向？

Yeah, I’m smiling as you said that, and the reason is that it comes back to my previous point as AI being this paradigm shift, and there are some customers who are really thinking on the bleeding edge about how to leverage the technology. We think AI can solve a plethora of business cases in productivity, content creation, creativity, and so on.
是的，听你这么说我在微笑，原因是这回到了我之前提到的人工智能作为一种范式转变的观点，有些客户确实在前沿思考如何利用这项技术。我们认为人工智能可以解决大量在生产力、内容创作、创造力等方面的商业案例。

So the way I’d answer that question is I think there’s a lot we can solve today rather than thinking through where it can be long term, but we believe there’s endless possibilities long term, even in the realm that you’re talking about.
所以我回答这个问题的方式是，我认为我们今天可以解决很多问题，而不是考虑它的长期发展，但我们相信在您所谈论的领域，长期来看有无尽的可能性。

Companies are often deciding between ChatGPT Enterprise and more traditional workplace productivity tools that have AI-infused capabilities, like Google Workspace or Microsoft 365 Copilot. How do you think about winning over companies deciding between those offerings?
公司通常在选择 ChatGPT Enterprise 和更传统的工作场所生产力工具（如 Google Workspace 或 Microsoft 365 Copilot，这些工具具有 AI 融合的功能）之间进行决策。您如何看待赢得在这些产品之间做出选择的公司的机会？

It doesn’t need to be an “or” statement. Ultimately we play well in a majority of environments. You mentioned Google—that’s definitely a product we play closely with, especially when you think about how to leverage ChatGPT. From my perspective, ChatGPT is the ultimate productivity tool, and it has a lot of horizontal use cases you could solve. I’ll give you an example about myself: A year ago I used ChatGPT for three distinct use cases: content creation, idea generation and data analysis. It’s gone well beyond that now. I use it as an assistant for strategy, for research, data analysis, so I’ve now increased my use cases. So I don’t know if it’s an either statement; it’s a complementary statement.
这不需要是一个“或”语句。最终，我们在大多数环境中表现良好。你提到了谷歌——这绝对是我们密切合作的产品，特别是当你考虑如何利用 ChatGPT 时。从我的角度来看，ChatGPT 是终极生产力工具，它有很多横向的使用案例可以解决。我给你一个关于我自己的例子：一年前，我使用 ChatGPT 进行三种不同的用例：内容创作、创意生成和数据分析。现在它已经远远超出了这一点。我将其用作战略、研究、数据分析的助手，因此我现在增加了我的使用案例。所以我不知道这是否是一个“要么”语句；这是一种互补的说法。

Incumbents like Microsoft, Google and AWS have pitched their cloud platforms as a safer place for enterprises to buy generative AI software than from newer startups like OpenAI, in part because they already have extensive contracts with large enterprises and are used to handling security and data compliance concerns. How are you trying to win over those customers?
像微软、谷歌和 AWS 这样的现有企业将他们的云平台宣传为企业购买生成性 AI 软件的更安全之地，而不是从像 OpenAI 这样的新兴初创公司购买，部分原因是他们已经与大型企业签订了广泛的合同，并且习惯于处理安全和数据合规问题。你们如何尝试赢得这些客户？

It’s twofold—one of our ultimate focuses is security. We want to be the most safe and secure platform out there, full stop. Security and safety internally is the No. 1 priority, and that translates to all of the compliance that we have. Security is an area that we will continue to mature into—we’re still a young company and our products are still young. ChatGPT Enterprise just turned one year old three months ago. If you look at our road map forward, security is a huge focus for us, and that’s how we represent it to our customers as well.
这有两个方面——我们最终的重点之一是安全。我们希望成为最安全、最可靠的平台，毫无疑问。内部的安全和保障是我们的首要任务，这也体现在我们所有的合规性上。安全是我们将继续发展的领域——我们仍然是一家年轻的公司，我们的产品也仍然年轻。ChatGPT Enterprise 三个月前刚满一岁。如果你查看我们的未来路线图，安全是我们关注的重点，这也是我们向客户展示的方式。

More broadly, when you’re talking to customers who are budgeting for generative AI, are they willing to spend heavily on the software, or are firms still cautious?
更广泛地说，当你与那些为生成性人工智能制定预算的客户交谈时，他们是否愿意在软件上大量花费，还是公司仍然保持谨慎？

This is one of my favorite questions. I would say 2023 was the year of experimentation and learning, and we were all trying to figure out the capabilities of what AI could do. In 2024 it started to look like, especially for enterprises, “How can this help me solve real use cases and add true business value?” And we’re seeing more and more of that, but an AI use case looked more like a project for an organization rather than something they were looking to scale.
这是我最喜欢的问题之一。我会说 2023 年是实验和学习的一年，我们都在努力弄清楚人工智能的能力。在 2024 年，尤其是对于企业来说，开始看起来像是“这如何帮助我解决实际用例并增加真正的商业价值？”我们看到这种情况越来越多，但人工智能的用例更像是一个组织的项目，而不是他们希望扩展的东西。

What we’re starting to see as we move towards 2025 is not just solving business value, but solving business value at scale. So what we’re hearing from our customers—in healthcare, manufacturing, pharma, legal, financial sector, the list goes on—they want to figure out how to take what is an AI-native product and transform their organization and do it at a scale that goes beyond just a project. They’re coming to OpenAI and saying, “Help us with our AI strategy from the bottom up.” I think that’s what the big change is going to be in 2025.
我们在迈向 2025 年时开始看到的不仅仅是解决商业价值，而是大规模地解决商业价值。因此，我们从客户那里听到的——在医疗保健、制造业、制药、法律、金融等行业，名单还在继续——他们想要弄清楚如何将一个原生于人工智能的产品转变为他们的组织，并以超越单一项目的规模来实现。他们来到 OpenAI，表示：“请从底层帮助我们制定人工智能战略。”我认为这就是 2025 年将会发生的重大变化。

I also wanted to ask you about distribution effects, both in the enterprise and with consumers. For example, with the Apple partnership that directs iPhone users to ChatGPT, have you modeled out how many iPhone customers you can convert to paying ChatGPT users?
我还想问你关于分配效应的问题，包括企业和消费者。例如，关于将 iPhone 用户引导到 ChatGPT 的苹果合作伙伴关系，你是否已经模拟出可以将多少 iPhone 客户转化为付费 ChatGPT 用户？

I can’t actually quote those numbers, but what I would say is that Apple is an incredible partnership for us, and it does offer a unique distribution mechanism for us, but also gives the power of OpenAI to a large number of users, which helps us achieve our mission.
我实际上无法引用这些数字，但我想说的是，苹果对我们来说是一个令人难以置信的合作伙伴，它为我们提供了独特的分销机制，同时也将 OpenAI 的力量赋予了大量用户，这帮助我们实现了我们的使命。

How are you thinking about market size for newer features like the o1 model, which seems like a power user feature because it takes up so much compute?
您如何看待像 o1 模型这样的新功能的市场规模，这似乎是一个强用户功能，因为它占用了如此多的计算资源？

Broadly, when we think about our models, with everything from GPT-4 to 4o to now our o-series models that are out, we’ve historically cut pricing as we get economies of scale for each of those models. And I imagine that will be the case in the future.
总体而言，当我们考虑我们的模型时，从 GPT-4 到 4o，再到现在发布的 o 系列模型，我们历史上在每个模型上随着规模经济的实现而降低价格。我想未来也会是这样的情况。

When you think about o1, it’s important to clarify that o1 is one of our latest milestones, and it can really reason through complex tasks and solve harder problems. For example, in law, it helps with legal or math calculations on term sheets, document analysis, multistep approval flows—it’s really incredible for legal. Another area is coding, which is something where it can help you build a full-stack application with very minimal prompting. Another key area is healthcare, think claim pricing, analyzing clinical guidelines, data analysis, the list goes on. And then manufacturing has been an incredible use case for it—that’s like mechanical design, optimizing production schedules, and different types of technical analysis. That’s where we’re seeing the most traction.
当你想到 o1 时，重要的是要澄清 o1 是我们最新的里程碑之一，它能够真正推理复杂任务并解决更难的问题。例如，在法律领域，它帮助进行条款清单上的法律或数学计算、文档分析、多步骤审批流程——在法律方面真的非常出色。另一个领域是编码，它可以帮助你在非常少的提示下构建一个全栈应用程序。另一个关键领域是医疗保健，考虑索赔定价、分析临床指南、数据分析，等等。制造业也是一个令人难以置信的应用案例——这涉及机械设计、优化生产计划和不同类型的技术分析。这是我们看到最多吸引力的地方。

We’ve heard that o1 pricing for the full model could be something like $2,000 per month given how advanced the model is. Would you say that’s in the realm of possibility?
我们听说完整模型的 o1 定价可能在每月约 2,000 美元，考虑到该模型的先进性。您认为这在可能范围内吗？

I’m not sure where that’s quoted from so I don’t have the context for that, but I would say our goal is to provide the best models at the best prices. And what we believe is that’s what our customers want, whether it’s in the enterprise or startups, and that’s where we’re going to continue to be competitive.
我不确定那是引用自哪里，所以我没有上下文，但我想说我们的目标是以最好的价格提供最好的模型。我们相信这正是我们的客户所希望的，无论是在企业还是初创公司，这就是我们将继续保持竞争力的地方。

Are you thinking about any other ways to monetize ChatGPT beyond subscriptions? Have you revisited the idea of adding advertisements or shopping links to ChatGPT?
你是否在考虑除了订阅之外的其他方式来实现 ChatGPT 的盈利？你是否重新考虑过在 ChatGPT 中添加广告或购物链接的想法？

For us, the focus is still on where we’re at today. We believe there’s a lot more to do for the current experiences that we have in ChatGPT and our API. Over time we’ll see if there’s the realm of possibility, I wouldn’t say to monetize, we actually just want to add value for users. Search is an opportunity for that. Another one is Canvas, which we recently launched. For now, it’s more about adding value than monetizing.
对我们来说，重点仍然是我们今天所处的位置。我们相信，在 ChatGPT 和我们的 API 中，还有很多工作要做。随着时间的推移，我们将看看是否有可能性，我不会说是为了盈利，我们实际上只是想为用户增加价值。搜索是一个机会。另一个是我们最近推出的 Canvas。目前，更重要的是增加价值而不是盈利。

There’s been a lot of talk lately about computer-using agents [which The Information has previously reported OpenAI is working on]. How are you thinking about competing in that realm, especially against players like Google that already have web browsers while OpenAI does not?
最近关于使用计算机的代理的讨论很多[《信息》之前报道过 OpenAI 正在研究这个]. 您是如何考虑在这个领域竞争的，特别是面对像谷歌这样的竞争对手，而谷歌已经拥有网络浏览器，而 OpenAI 则没有?

Currently there’s not too much broad conversations we’re having externally on agents. It is something that we believe is the next frontier as AGI [artificial general intelligence, or humanlike intelligence]. What we’re focusing on today is reasoning, as I mentioned with the o1 models, but over time you will see us start to veer into the agents category. We believe we’ve started to solve the first step of that with o1, and agents are the next step for us.
目前我们在外部关于代理的广泛讨论并不多。这是我们认为的下一个前沿领域，即 AGI（人工通用智能或类人智能）。正如我提到的，我们今天关注的是推理，使用的是 o1 模型，但随着时间的推移，您将看到我们开始转向代理类别。我们相信我们已经通过 o1 解决了第一步，而代理是我们接下来的步骤。

More broadly, what do you see as the next wave of capabilities or potential use cases for customers that will come online in the coming year or so?
更广泛地说，您认为在未来一年左右，客户将上线的下一波能力或潜在用例是什么？

When we think about the levels of AGI, each level we believe will solve new use cases. The first level was the chatbot, the second level was reasoning, the third level is agents, and the fourth level is what we call innovators [AI that can generate groundbreaking ideas], and then level five is [AI that can function as] organizations. And when you think about level five, that’s when we believe we’ll be at full AGI.
当我们考虑 AGI 的各个层级时，我们认为每个层级将解决新的用例。第一个层级是聊天机器人，第二个层级是推理，第三个层级是代理，第四个层级是我们所称的创新者[能够产生突破性想法的 AI]，然后第五个层级是[能够作为]组织的 AI。当你想到第五个层级时，我们相信那时我们将达到完全的 AGI。

So for use cases, the first level is basic conversational AI systems similar to the current chatbots we have today, ChatGPT being the leader. The second is reasoning, which we’ve just talked about, that gives you the capability to solve really complex problems, and you need reasoning abilities to do that, and we talked about cases across legal or manufacturing. Now, the third level, what you’re referring to as agents, these are systems that can autonomously take actions and complete a task, and that’s something that will open up a lot of new range of use cases, and it’s something that is going to be possible very soon.
因此，在使用案例中，第一级是基本的对话式人工智能系统，类似于我们今天所拥有的当前聊天机器人，以 ChatGPT 为首。第二级是推理，我们刚刚讨论过，这使您能够解决非常复杂的问题，而您需要推理能力来做到这一点，我们讨论了法律或制造等领域的案例。现在，第三个级别，您所提到的代理，这些是能够自主采取行动并完成任务的系统，这将开启许多新的使用案例，并且这很快就会成为可能。