There is a deeply held belief in education (and in leadership more broadly) that if we can make things more efficient, everything will get better.
We will save time. We will reduce costs. We will ease the burden on people. And in doing so, we will create space for what matters most.
Artificial intelligence has only strengthened that belief. It promises to streamline lesson planning, automate communication, accelerate data analysis, and reduce administrative load across our systems. It is easy, then, to assume that AI will simplify the work of schooling.
But history suggests something more complicated and perhaps more consequential.
In the 19th century, the economist William Stanley Jevons observed what has come to be known as Jevons paradox.
As steam engines became more efficient, the expectation was that coal consumption would decline. After all, if each engine used less coal, total demand should fall.
Instead, the opposite occurred. Coal consumption increased dramatically.
Greater efficiency reduced the cost of using steam power, which made it viable in more places and for more purposes. What began as a technological improvement became an expansion of the entire system.
The lesson is subtle but important. Efficiency does not always lead to reduction. Often, it leads to expansion.
I believe that dynamic is beginning to emerge in education as well.
AI in schools, and efficiency
Everywhere we look, AI in schools is lowering the cost of cognitive work. Tasks that once required significant time and effort can now be completed in seconds.
A teacher can generate multiple versions of a lesson. A principal can produce a detailed report with minimal effort. A central office team can analyze large sets of data and communicate insights almost instantly.
If we follow the logic of efficiency alone, we might expect this to reduce workload and create breathing room across the system. This seems to be the thing everyone is saying, but I believe we might have it wrong.
Because systems rarely respond that way.
When something becomes easier to produce, we tend to produce more of it. When something becomes faster, we expect it more frequently. When something becomes cheaper, we expand its use.
Consider instructional practice. If AI makes it easier to differentiate learning, the natural inclination will not be to maintain current expectations while working less.
Instead, expectations will rise. What was once considered exceptionalhighly personalized instruction, frequent feedback, multiple pathways through contentbecomes standard.
The ceiling moves. The system adjusts. And the overall volume of work expands, even as the effort required for any single task decreases.
A similar pattern is likely to emerge in administrative work. Reporting, compliance, and communication may all become more efficient, but that efficiency will invite new demands.
Districts may find themselves producing more frequent updates, more detailed analyses, and more tailored communications to stakeholders. The time saved in one area is quickly reinvested into expanding the scope of the work.
This is where the implications move beyond the classroom and into the broader operation of school systems.
Take school construction, for example. A surface-level interpretation of AI might suggest a reduced need for physical space. If more learning can occur virtually, perhaps we will need fewer buildings or smaller campuses.
But viewed through the lens of expansion, a different possibility emerges. If AI enables districts to offer more diverse and personalized learning experiences, then the demand for specialized spaces may increase.
Career and technical education programs, innovation labs, collaborative environments, and student support centers could all expand in response to a system that is now capable of offering more. The question may not be whether we need less space, but whether we need different space, designed for a more complex and flexible educational model.
The same tension appears in funding. Efficiency is often equated with cost savings, but that relationship is not guaranteed.
If AI increases the number of programs, raises expectations for personalization, and expands the scope of services provided to students and families, total system costs may rise even as the cost of individual tasks declines. Districts could find themselves more efficient in execution, but operating within a larger and more demanding system.
Staffing presents a similarly nuanced challenge. It is tempting to assume that greater efficiency will lead to fewer roles.
In practice, it is more likely to lead to different roles. As routine tasks are automated, the value of human work shifts toward judgment, relationships, and strategic thinking.
New responsibilities emerge even as others recede. The system does not necessarily contract. It evolves.
What not to do next
At its core, the Jevons paradox is not really about technology. It is about behavior within systems.
When constraints are removed, when something becomes easier, faster or more accessible, we tend to expand our use of it. That instinct is not inherently problematic; in many ways, it is what drives progress. But it does create a challenge for leaders.
Without intentional discipline, systems will grow in complexity faster than they grow in coherence.
That may be the central leadership question in the age of artificial intelligence. Not whether AI will make us more efficientit will. Not whether it will increase our capacityit already has.
The question is whether we will pair that capacity with clarity about what we are trying to achieve and restraint about what we choose to take on.
Because in a world where we can do more than ever before, the limiting factor is no longer capability; it is focus.
And the systems that navigate this moment most effectively will not be those that adopt AI the fastest, but those that are most intentional about what they allow their systems to become.
In the end, better tools do not simplify systems on their own. They expand what is possible. And leadership, more than ever, is the discipline of deciding what not to do next.
