Ling-yi Tsai stands at the forefront of the HRTech revolution, bringing decades of experience in guiding organizations through the complex terrain of technological change. Her expertise in HR analytics and talent management has made her a critical voice for companies trying to balance the efficiency of artificial intelligence with the long-term health of their human capital. As we move into an era where AI can mimic professional output with startling speed, she argues that we are facing a hidden crisis of intellectual erosion. This conversation explores how the decline in junior hiring and the over-reliance on automated tools are hollowing out the very expertise that makes AI useful in the first place, and what leaders must do to ensure their teams do not lose the ability to think critically in an increasingly automated world.
The discussion addresses the alarming decline in entry-level tech roles and why the “tech-savvy” younger generation is actually at the highest risk of losing professional judgment. We explore the limitations of current AI adoption metrics, the necessity of treating AI as a “sparring partner” rather than a shortcut, and the practical ways organizations can use “slow” rituals like architecture-inspired studio models to build the tacit knowledge that algorithms cannot replicate.
Entry-level hiring in tech-heavy sectors has dropped significantly in recent years, potentially stalling the growth of the next generation of experts. How do you view the long-term impact of this “entry-level freeze” on an organization’s internal talent pipeline?
The numbers we are seeing are quite startling and suggest a structural shift that could haunt organizations for decades. Entry-level postings in U.S. roles exposed to AI dropped by a massive 35% between January 2023 and June 2025, which tells us that the “entry-level” gate is closing just as the technology is peaking. Even more concerning is that employment for developers specifically aged 22 to 25 has declined nearly 20% from its 2022 peak, while older, more experienced workers in those same fields are actually seeing growth. We are essentially cutting off the bottom of the pyramid, which means the tacit knowledge built only through years of hands-on, messy problem-solving is not being passed down. If companies stop training juniors because they think AI can do the basic work, they will eventually find themselves with a leadership layer that has no foundational experience to draw upon when the AI fails.
There is a common assumption that younger, tech-native employees are best positioned to leverage AI, but research suggests domain expertise is actually the deciding factor in its effectiveness. Why does an experienced professional often outperform a “power user” when working alongside these models?
It is a common misconception that fluency with the tool equates to high-quality output, but the data proves that experience beats tech fluency every time. Research in Management Science has shown that AI complements domain expertise rather than replacing it, meaning the tool creates the most value when the person using it already understands the nuances of the problem. A Harvard Business School study even highlighted that workers with relevant expertise could spot subtle gaps in AI output and fill them with professional judgment, whereas those further from the domain simply could not match that quality, even with the same model. We have to remember that AI is excellent at codified knowledge, but it struggles to replicate the “gut feeling” or tacit knowledge that senior professionals have spent years refining. Without that human control mechanism, the AI’s output is often just a very confident-looking version of a mediocre or incorrect answer.
Most leadership teams are currently obsessed with adoption rates and query volume as signs of success. What are the dangers of these shallow metrics, and how should we actually be measuring the health of our intellectual capital?
The danger is that we are measuring “how much” people use the tool instead of “how well” they are thinking, which are two very different things. In a Microsoft and Carnegie Mellon study of 319 knowledge workers, researchers found a troubling trend: the more confidence workers placed in AI, the less critical thinking they actually applied to checking the results. This is backed by a separate study of 666 participants published in the journal Societies, which found a negative correlation between frequent AI use and critical thinking, largely because people were offloading their reasoning to the machine. On a standard corporate dashboard, an employee who runs fifty unchecked queries looks identical to one who runs five after carefully working the problem first. To protect our intellectual capital, we need to shift our metrics to focus on whether an employee can explain why an output is correct, perhaps by requiring them to document their own reasoning before they even touch the AI.
You have advocated for using AI as a “sparring partner” rather than a shortcut. How can managers practically restructure daily workflows to ensure their teams are sharpening their judgment instead of letting it atrophy?
We need to treat AI as a tool that challenges us rather than a machine that does the work for us, which requires a very deliberate change in workflow habits. For example, empirical research into how experienced accountants use AI found that they used the model’s confidence scores to target their own review efforts, using their professional judgment as a control mechanism rather than blindly accepting the result. Managers should encourage junior analysts to draft their own recommendations or solve a problem on a blank sheet of paper before they ever consult the AI. Once they have their own answer, they can query the AI and then be required to explain any divergence between their logic and the machine’s logic out loud. The goal here isn’t to get to the answer faster; it’s to force the human to get sharper at spotting the precise moments where the AI struggles, which is a skill that machines simply cannot replicate on their own.
Speed is the primary currency of the AI era, yet you suggest that “slowness” is actually a vital component of building expertise. How can firms maintain a rigorous problem-solving culture when the temptation to produce “finished-looking” work instantly is so high?
The “finished” look of AI output is its most dangerous trait because it encourages us to skip the rigorous, often slow, process of debate and peer review. Research from Microsoft and CMU showed that standards slip first on the tasks that matter least to performance reviews, meaning the erosion of quality starts exactly where nobody is looking. To fight this, firms must preserve the rituals that built judgment before these tools existed, such as case debates without any AI assistance or brainstorming sessions that start with nothing but a blank page. I’m a big advocate of the “studio model” borrowed from architecture, where junior staff present unfinished, raw reasoning to senior colleagues who then walk them through how they would personally approach the problem. This type of apprenticeship is undeniably slower than asking an AI for a draft, but that very slowness is what builds the deep expertise that will keep a company resilient when the model encounters a problem it has never seen before.
What is your forecast for the future of organizational expertise in the next decade?
In a decade, we will likely see a sharp divide between organizations that treated AI as a shortcut and those that treated it as a cognitive enhancer. Just as the Industrial Revolution eventually led to a measurable public health crisis like obesity because we stopped moving our bodies, we are now facing a period of “cognitive atrophy” because we are stopping the effortful part of thinking. My forecast is that companies ignoring the warning signs today will find themselves in ten years with a structural vacancy of talent, realizing too late that they have no one left in the building who understands the “why” behind the “what.” The leaders who win will be the ones who recognize that AI is a tool for experts, and that the only way to create those experts is to allow them the space to struggle with problems long before they reach for the technology.