In the AI era, the difference between high-potential talents and ordinary employees lies in three capabilities

A study has found that AI is reshaping employers' skill expectations for new hires. Future professionals must possess three core capabilities: taking on multiple roles independently, integrating knowledge from diverse sources, and redesigning workflows with AI. These three capabilities are emerging as the new benchmark that distinguishes high-potential talent in the AI era.

"Three years ago, no one in my department used generative AI tools. Today, 100% of the team uses them every single day, often for several hours at a time," an executive at a large multinational bank recently told us. His observation reflects the explosive growth of generative AI adoption in the modern workplace.

We studied 30 organizations across three industries — banking and finance, management consulting, and technology (including media and e-commerce) — sectors that employ the largest share of MBA graduates, to examine how the rise of generative AI has transformed employers' skill requirements for new recruits. Below are the three essential capabilities professionals must master to land jobs and thrive in the AI era.

Taking on Multiple Roles Independently

AI frees up professionals from spending time on low-level tasks, while empowering them to handle multiple responsibilities on their own. The transformation of product management in software technology companies perfectly illustrates this shift.

Traditionally, products are built by the "iron triangle" of product managers, user experience designers, and developers, with each professional specializing in one role throughout their career. But AI is breaking down the boundaries between these roles. Fundamental product management tasks — tracking work orders, coordinating workflows, compiling data, conducting analysis, and writing reports — are now being automated by AI. AI can also generate user interface designs directly from prompts, gradually replacing entry-level prototyping work. As a result, the executives we interviewed predict that junior roles in product management, user experience, and engineering will merge into a single "universal technical specialist" position, where practitioners use AI tools to create prototypes, write initial code drafts, and deliver finished products.

With product managers spending less time on basic execution tasks, software companies have raised their expectations for them: they are now expected to be more strategic and long-term oriented. Future product managers must not only understand business operations but also grasp the technical aspects of products; they need to fully master the end-to-end product lifecycle, map out clear roadmaps, and anticipate potential challenges along the way.

Deep professional expertise in a specific field — such as engineering, finance, or statistics — remains indispensable, but it is no longer sufficient on its own. Professionals must also develop system-level thinking to understand how different components of a product integrate to create value. They need to leverage AI tools like Claude, Figma, Lovable, and GitHub Copilot to move from initial ideas to testable prototypes that verify the product delivers the right experience for target users, and then build working proof-of-concepts to confirm technical feasibility. In other words, they must adopt an "AI-first" problem-solving mindset to iterate faster and develop better product concepts and prototypes.

The example of building an AI customer service agent for an online marketplace clearly demonstrates this logic. Before the advent of AI, human agents handled customer service, and the core function of the department was people management — training, setting performance metrics, monitoring results, and delivering feedback. But when AI takes over customer service, the core function shifts to product management. AI has rewritten the rules of the game: customer service is now embedded directly into the product or service itself, allowing customers to resolve issues that previously required human support, or even complete tasks that were impossible before.

This means product managers must fully understand customer needs to design end-to-end solutions. They need to figure out: What data is required to train the AI agent? What instructions and specifications should be defined? What AI agents need to be deployed and how should they be arranged? How can feedback from customers and frontline teams be incorporated to continuously improve performance?

Integrating Knowledge from Diverse Sources

The transformation of new product development in the manufacturing industry best exemplifies this capability. Developing new products is a complex endeavor that requires massive investment and deep market expertise across multiple domains, including product design and engineering, supply chain, sustainability, finance, and user experience design. A team of specialists would spend months meticulously collecting information, analyzing data, and executing various activities to design a single new product.

A top-tier management consulting firm showed us how AI-powered product development can accelerate physical product innovation. A dedicated set of AI tools can mine internal customer data to generate and test concepts, estimate product costs based on product specifications and historical cost data, rapidly validate ideas using synthetic customer personas that represent target user groups, and evaluate different procurement and transportation options.

These AI agents are programmed to extract valuable insights from unstructured data scattered across different departments of an organization. For example, AI can quickly generate a list of suppliers that meet specific sustainability requirements. Customer feedback from previous product versions can also be used to test new concepts. Neural networks are replacing traditional physical modeling and mathematical solvers as new design tools.

In this context, the most critical responsibility of new product designers is not technical design itself, but rigorously controlling the quality of input information, building test scenarios, stress-testing AI outputs, and refining and fine-tuning final designs. They need a comprehensive understanding of data across their organization and supply chain, identify key performance drivers, and pinpoint gaps in their understanding of customers. For instance, they must ensure synthetic user personas fully reflect the diversity of their customer base.

Redesigning Workflows with Embedded AI

When AI agent tools are deployed to automate tasks like data collection, analysis, and reporting, the time required for these activities drops to nearly zero. In their place, new tasks emerge that require human execution, such as training and supervising AI models. To maximize value from AI in this new environment, managers must not only integrate AI tools into existing workflows, but also design entirely new workflows from scratch.

This process begins by identifying which tasks will be automated by AI, which tasks still require human involvement, and what new capabilities need to be added. Similar to process design in manufacturing plants, managers should assess the required resources, training, instructions, specifications, and information for every single task. Work rhythms also need to be adjusted, as end-to-end cycles are drastically shortened: project management meetings are reduced, tasks that previously took weeks can now be completed in less than a day with far fewer resources.

Katherine Zhao, Head of API Products at OKX, told us that agentic AI has cut the API release cycle at her fintech company by 30% to 60%. In the new workflow, the role of product managers has evolved. Instead of spending time searching for information to meet user needs, filling gaps in software specifications, and designing test cases, they now focus on defining product requirements, developing concepts, evaluating the feasibility of agentic AI solutions, and assessing the adequacy of edge case testing and release risk control.

In addition, a new task has been added to workflows: overseeing AI operations. When AI is authorized to deploy products, even minor errors can escalate into significant production risks. To minimize these risks, managers must establish clear governance frameworks rather than prioritizing efficiency alone.

Executives at the companies we studied made it clear that they are investing in training existing mid-level professionals to become AI-proficient, rather than replacing them with new hires who know how to use AI. They emphasize that domain expertise, decision-making abilities, and accumulated experience of these professionals remain irreplaceable. For MBAs seeking jobs in the three industries, the bar has been raised. Executives expect them to not only have a solid foundational knowledge of business, but also proven hands-on experience applying AI tools to solve business problems and using critical thinking to verify the reliability of AI outputs.

Keywords: #TalentManagement

Jim Doucette and Vishal Gaur | Text

Jim Doucette is a retired partner of EY-Parthenon and McKinsey & Company. Vishal Gaur is a Professor of Operations, Technology and Information Management at the SC Johnson College of Business, Cornell University.

Qiang Zhou | Editing & Proofreading

This article is republished with permission from the WeChat Official Account "Harvard Business Review" (ID: hbrchinese), authored by HBR-China, and authorized for distribution by 36Kr.