Over the last few years, I’ve studied this shift in the workplace firsthand, both in large-scale data analysis of how software developers are using AI tools and as a consultant to multinational corporations rolling out AI agents. The evidence is clear: the primary opportunity AI provides is not to replace people, but to reallocate their focus. My research—co-authored with colleagues from UC Irvine, Microsoft, GitHub, and the Linux Foundation—has shown that generative AI excels at absorbing the administrative tasks that bog down employees, freeing them to concentrate on the creative and complex work that truly moves the needle.

But that efficiency has led leaders to a crucial strategic choice — one that I believe many organizations are getting wrong. Do you reinvest the freed-up capacity of your human employees into innovation and growth, or do you cash it in for short-term cost savings? We’re already seeing companies choose the latter, often by cutting entry-level positions typically tasked with routine, lower-level work. On a surface level, the logic is tempting: why pay a salary for tasks an algorithm can do for pennies on the dollar?

This is a profound strategic error.

Junior employees are typically innovative and technically adept, and in tune with a new generation of customers. More importantly, they become tomorrow’s managers and leaders. Cutting them off not only silences crucial perspectives but also creates a long-term deficit in institutional knowledge, breaking the chain of skills that develops as employees grow within a company. This approach is also counterproductive because our research shows that it’s actually individuals with lower experience who benefit the most from artificial intelligence tools, which can help them build new capabilities and manage complex tasks. The path from entry-level employee to valued contributor might actually be shorter than ever.

That’s why the smart, long-term play isn’t to cut employees with “lesser” skills, but to complement them. For managers, the first step is to lean into mentorship by partnering early-career employees with seasoned professionals. Both parties benefit: the junior employee brings a fresh perspective (and often, a higher level of comfort with new technology), while the veteran provides institutional knowledge and critical insights on professional norms and best practices.

This approach also presents a powerful strategic opportunity. While your competitors are under-investing in their future, you can attract and hire the best of the next generation. And thanks to AI’s ability to reduce the burden of project management and administrative tasks, those highly skilled professionals will have more bandwidth available to support them.

But the management challenge doesn’t stop there. While generative AI helps employees learn faster and test their ideas, its speed and reliability are a double-edged sword that can cut apart the connective tissue of an organization. The quick, informal query that was once directed to a peer in the next cubicle is now sent straight to a chatbot. The difficult conversations inherent to collaboration can be avoided — but employees don’t get the chance to work through a challenge and grow their communication and problem-solving skills. If left unchecked, organizational siloes will grow and innovation will suffer.

This reality creates a new, urgent mandate for management. The primary challenge of the AI era is not technological, but organizational.  The time and profit dividends AI provides must be consciously reinvested in building and strengthening interpersonal connections, particularly between junior and senior talent, and in fostering a corporate culture that rewards the kind of truly creative thinking that AI can’t emulate. The real measure of leadership in the coming years will not be how quickly companies adopt AI, but how wisely they invest in the people who use it.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.