内容来源：https://aibusiness.com/generative-ai/ai-agents-supercharge-business-model

内容总结：

谷歌云赞助报道：企业如何选择首个生成式AI应用场景——从解决具体问题入手

纽约讯——在人工智能代理大会（AI Agent Conference）上，多家企业高管与行业专家指出，企业启动生成式AI应用的关键在于聚焦能改善“人与信息交互体验”的领域，而非盲目追求技术前沿。

从纸质股票到AI驱动的资本平台

股权管理公司Carta的首席产品官普鲁莎莉·鲍尼卡分享了一个成功转型案例。Carta最初以纸质股票凭证数字化起家，一度难以吸引投资者关注。但如今，该公司已发展为年收入近6亿美元的私人资本平台，其秘诀在于将AI注入业务模式。

鲍尼卡表示，公司的新战略是：从服务业务起步，将其转化为AI驱动的产品业务，再实现规模化扩张。“全球最大的商业机会，就是把过时的服务业务，变成由AI赋能的创新型产品业务。”

小处着手，亲身体验

对于仍处于摸索阶段的企业，鲍尼卡建议“选择一个非常具体的问题进行实验”。她透露，Carta的AI学习进程得益于早期为Anthropic公司的Claude模型开发命令行界面、插件和技能，这帮助他们深入理解了代理的行为特性。

全球教育测评公司培生的首席技术官大卫·特里特强调，尤其是企业高管必须“亲自上手”，才能真正理解AI的力量或潜力。他认为，当高层管理者掌握AI工具的使用方法时，企业的变革效果将截然不同。

避免“工具焦虑”，重在流程再造

数字化转型公司Apply Digital的首席执行官阿里·阿尔卡法吉提醒企业，不应将AI自动化视为额外工作负担，而应“重新构想整个工作流程”，找到AI真正能颠覆工作方式的地方。

美国团购平台Groupon的商户体验副总裁玛莎·夏尔玛则观察到一种“工具焦虑”现象：“坦率地说，我看到很多人因为试图紧跟所有技术前沿而不知所措。”她建议企业“放慢脚步”，避免因害怕被时代抛弃而盲目试错。

正视失败：建立安全护栏与保留人类监督

多位高管承认，AI转型必然伴随失败。Carta也曾有过“灾难性”的实验——例如其直接操作数据的尝试以失败告终。目前，Carta禁止代理直接访问数据，仅允许其通过包含数据健康检查的工作流程使用产品。

阿尔卡法吉预测，未来AI市场势必会出现重大事故，这可能让行业“三思而行”，但也将推动更高标准的安全规范建立。他所在的公司已制定了一套原则和指南，用于在代理出错时实时决策。

Sunrise AI的首席执行官迪帕克·斯里瓦斯塔瓦强调，建立人类信任的关键在于“人在回路中”的监督机制。当AI代理被授权进行金融交易或购物决策时，人类参与是维持这种信任的最佳方式。

夏尔玛进一步指出，“人”的因素也是企业实现差异化竞争的核心。她认为，随着AI代理输出的内容趋于同质化，那些希望提供个性化产品和服务的企业，必须保留人类参与，以创造独特价值。

中文翻译：

由谷歌云赞助
选择您的首个生成式AI用例
要开始使用生成式AI，首先应聚焦于能够改善人类与信息交互体验的领域。

股权管理公司Carta是企业如何运用AI与智能体来重塑和塑造业务的范例。

纽约——Carta最初以纸质股票证书数字化而闻名，一度难以获得投资者关注。但多年来，它已发展成为私募资本平台，年收入近6亿美元，其秘诀在于近期重新调整并融入AI技术的运营模式。

Carta首席产品官Vrushali Paunikar在周一的AI智能体大会上表示，公司最初的战略是推动业务快速增长：从服务业务起步，将其转化为软件，占领市场，再重复这一模式。这家股权管理公司凭借这一策略取得了成功，而AI技术与智能体的发展则进一步放大了其成果。

“当今世界最大的商业机遇，就是将过时的服务业务转变为由AI驱动的产品业务，”Paunikar在演讲中表示。她指出，新的业务战略始于服务业务，随后将其转化为AI赋能的产品业务，最终实现规模化。

Carta的精细化商业模式，正是企业为融合生成式AI与智能体AI而不得不进行转型的典型案例。尽管这家股权公司在AI应用上取得了成功，但一些企业仍在探索这项技术在其组织中的最佳落点。

对于仍在摸索如何运用AI的企业，Paunikar建议从小处着手。

“选择一个非常具体的问题进行实验，”她在采访中表示。她补充道，加速Carta学习智能体应用的关键之一，是使用Anthropic公司的Claude模型。

“我们开始为Claude构建命令行界面、插件和技能，”她说，“这实际上帮助我们深入了解了智能体的行为模式。”

教育评估公司Pearson的全球首席技术官David Treat认为，尝试AI工具对企业，尤其是高管层而言至关重要。

“你必须亲自动手，才能真正理解其力量或潜力，”Treat在炉边谈话中表示，并补充道，当高管们掌握AI工具的使用方法时，效果会截然不同。

数字转型公司Apply Digital的CEO Ali Alkhafaji则表示，尽管实践AI工具至关重要，但企业不应将AI自动化视为工作流程的简单叠加。

“重新构想流程和工作负荷，”他说，“我敢保证，大多数情况下，你会发现AI能帮你真正转变工作方式。”

然而，Groupon商户体验副总裁Masha Sharma在采访中指出，企业需要避免操之过急，或认为必须尝试所有AI工具；它们需要保持灵活，并能够快速构建下一个最佳方案的雏形。

“坦白说，我看到很多人因为试图紧跟所有前沿技术、担心自己落伍而感到不堪重负，”她说，“你需要适当放慢脚步。”

不过，业务领域的AI转型并不意味着没有失败和错误。尝试生成式AI与智能体AI的企业应认识到，它们将面临有时无法预见的挑战。

“为AI制定所有规则和政策以保护企业极其困难，因为你不知道潜在的事故会从何而来，”Alkhafaji在采访中表示。

即便是商业模式已精细化的Carta，在成功之前也曾经历过失败实验。

“我们曾尝试直接操控，结果一败涂地，”Paunikar说。她补充道，Carta不会让智能体直接访问其数据。智能体只能通过带有数据健康检查和验证的工作流程来访问产品。

为了应对这些未知变量，Alkhafaji表示，Applied Digital已建立了一套原则和指导方针，当智能体出现错误时，公司可在实时决策中应用这些准则。

“我们很可能在某个时刻面临挑战，”他谈到AI市场时补充道，“事故总会发生，且可能涉及一个主要品牌。这会让许多人对AI三思而行。我不认为这会阻止AI发展，但无疑会提出更高的监管标准——这些标准当前急需，却并未得到普遍采纳。”

在实验阶段避免灾难性后果的一种方法是保持人工介入。

“人类的核心价值之一是信任，”Sunrise AI的CEO Deepak Shrivastava在采访中表示。他补充道，随着智能体被信任去执行金融交易、做出购买决策甚至进行购物，这种信任度正在提升。“建立并维护信任的最佳方式是人与人之间的直接互动。”

Sharma表示，人工介入不仅是为了维护信任，更是企业实现差异化竞争的关键。

“一切都会开始变得非常相似，”她谈到AI智能体的输出时表示。她补充道，想要脱颖而出、提供个性化产品或服务的企业，仍需人类深度参与。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Equity management firm Carta is an example of how enterprises can use AI and agents to transform and shape their businesses.
NEW YORK -- Carta began as a company known for digitizing paper stock certificates, struggling to get investors' attention. Still, over the years, it has evolved into a platform for private capital, with annual revenue of nearly $600 million, using a playbook that the company recently revamped and is now powered by AI.
Carta’s original strategy was to supercharge its business: Start with a service business, transform it into software, dominate the market and repeat, said Vrushali Paunikar, the company's chief product officer, during a presentation at the AI Agent Conference on Monday. While the equity management firm has found success with this approach, the growth of AI technology and agents has helped it magnify results.
“The greatest business opportunity out there in the world is taking a dated service business and turning it into a product business powered with AI,” Paunikar said during the presentation. She said the new business strategy starts with a service business, transforms it into an AI-enabled product business and then scale.
Carta’s refined business methodology is an example of how businesses are having to shift to incorporate generative and agentic AI. While the equity firm has found success with AI, some enterprises are still figuring out where the technology fits within their organizations.
For enterprises still trying to understand how to use AI, Paunikar advised starting small.
“Pick a very finite problem and experiment,” she said in an interview. She added that one of the things that accelerated Carta’s learning process with agents is using Claude from Anthropic.
“We started building like CLIs and plugins and skills for Claude to use,” she said. “That actually helped us learn a lot about agent behavior.”
Experimenting with AI tools is the key for enterprises, especially C-level executives, according to David Treat, global CTO at Pearson, an education and academic assessment company.
“You have to be hands-on to really understand the power or potential,” Treat said during a fireside chat, adding that when C-suite executives know how to work with AI tools, it makes a difference.
While practicing with AI tools is essential, enterprises should not view AI automation as just another layer to add to their workload, said Ali Alkhafaji, CEO of Apply Digital, a digital transformation company.
“Reimagine that process, that workload,” he said. “I guarantee you more often than not, you’ll find places where AI can help you truly transform the way you work.”
However, businesses need to avoid rushing or thinking they need to experiment with all AI tools; they need to be nimble and able to prototype the next best thing, said Masha Sharma, vice president of merchant experience at Groupon, in an interview.
“Frankly, I see people getting overwhelmed because you’re trying to be on the cutting edge of all of that and you’re thinking that you’re missing out,” she said. “You kind of have to slow down.”
However, AI transformation in business does not mean no failures and mistakes, and enterprises experimenting with generative and agentic AI should know that they will face challenges that sometimes cannot be predetermined.
“It is incredibly difficult to put all the rules and policies in place to protect enterprises with AI because you don’t know where a potential incident is going to come from,” Alkhafaji said in an interview.
Even Carta, with its refined business model, failed some experiments before succeeding.
“There were some experiments we did on direct manipulation, It was disastrous,” Paunikar said. She added that Carta does not give agents access to its data. Agents can access its product only through workflows that have data health checks and validation.
To deal with these unknown variables, Alkhafaji said Applied Digital has set up a set of principles and guidelines that can be used to make decisions in real time within the company if an agent makes a mistake.
“We’re probably going to face a challenger moment at some point,” he added, referring to the AI market. “An incident will come up, and it is going to be a major brand. It is going to make many people think twice about AI. I do not think it is going to stop it, but it is certainly going to put up like an additional level of rigor that is needed today but not really adopted everywhere.”
One way to avoid a disastrous moment during the experimentation phase is to keep a human in the loop.
“Where do you place value in general as a human is trust,” said Deepak Shrivastava, CEO of Sunrise AI, in an interview. He added that, with agents being trusted to make financial transactions, make purchasing decisions or even shop, that level of trust is increasing. “The best way to build that trust and maintain that trust is people, human to human.”
More than maintaining trust, the human-in-the-loop, or people factor, is also a way for enterprises to differentiate, Sharma said.
“Everything is going to start to look very much the same,” she said, referring to the responses from AI agents. She added that businesses that want to stand out and personalize their products or services will need humans to stay involved.