When the Teacher Can Breathe Again

This week, Education Week published a study from researchers David Marshall and Tim Pressley with a finding almost no one has been talking about.¹ Most conversations about AI in classrooms focus on what AI does to students. This one looked at what AI does to teachers.

The result is quiet, but worth pausing over. Teachers with higher confidence in selecting and integrating AI tools (what the researchers call "AI pedagogy self-efficacy") reported stronger self-efficacy for engaging students and managing instruction. That stronger sense of capability was linked to lower perceived workload. The lower workload predicted lower anxiety. The lower anxiety predicted better mental well-being.¹

It is a chain of effects that runs in the opposite direction from the burnout story we usually hear. The popular narrative holds that technology piles new demands on teachers: new tools to learn, new policies to enforce, new student behaviors to police. The Marshall and Pressley findings suggest something different can happen when AI is integrated well. Teachers do not just get more done. They feel more in control of what they are doing.

That feeling of control, it turns out, may be the most underrated variable in education technology.

What the Burnout Numbers Do Not Show

For nearly a decade, teacher burnout has been measured in workload hours, class sizes, paperwork volumes, parent emails. The interventions have followed: more planning periods, fewer required forms, software that promises to save time. None of it has bent the trend lines very much. In a January 2025 California Teachers Association survey, 40% of teachers said they were considering leaving education within the next few years.²

The Marshall and Pressley study points at something subtler. The burden teachers feel is not just the volume of work. It is the sensation that the work is happening to them, rather than being directed by them. A teacher who watches a student struggle and does not know why, or who grades thirty essays without seeing how each one was actually built, accumulates a quiet kind of fatigue. It is the fatigue of operating on hidden information.

When AI is introduced poorly, when it generates content the teacher cannot understand or makes recommendations the teacher cannot interpret, it deepens that fatigue. When it is introduced well, it can reverse it. Teachers begin to see what they could not see before. A class becomes legible. The day becomes shapeable.

What "Used Well" Actually Looks Like

The study does not prescribe a particular tool. It identifies a condition: teachers must feel capable. They must understand what the AI is doing, why it is doing it, and how to redirect it. They must be able to trust the outputs, and trust their own judgment about those outputs.

This is where most enterprise AI in classrooms still falls short. A black-box assistant that hands a teacher a finished worksheet does not increase their sense of control. It substitutes for their judgment without explaining itself. A dashboard that surfaces metrics without showing the underlying student work asks the teacher to trust the system's interpretation rather than form their own.

The form of AI that builds teacher confidence, by contrast, is the form that makes student thinking visible. Not just the answer, but the path to the answer. Not just the score, but the revisions, the pauses, the moment a student deleted a sentence and tried again. When a teacher can see how learning actually unfolded, they stop having to guess. And the absence of guessing is, quietly, the absence of one of the most exhausting parts of the job.

The Capability That Comes From Clarity

It is easy to misread the Marshall and Pressley result as a story about training. Give teachers more PD on AI tools, the logic might go, and self-efficacy will rise. There is some truth in that. But the deeper reading is structural. Self-efficacy is not just a function of how much a teacher knows about a tool. It is a function of whether the tool gives them something they can act on.

A confident teacher is one who can answer the basic questions of their craft. Where is each student right now? What did they try before they got stuck? Who has gone quiet, and for how long? These are not metric questions. They are texture questions, and most current education technology does not answer them at all. It tells you who scored well and who did not. The teacher is left to fill in the rest from memory and intuition, often across more than a hundred students a week.

The capability the study identifies is not a feature of any particular AI product. It is a property of the relationship between the teacher and the information their classroom produces. When that information is rich, honest, and human-readable, the teacher's sense of agency grows. When it is thin or opaque, the sense of agency collapses, no matter how powerful the underlying technology.

The Quiet Promise

This is the heart of what we are building at Koan. The WorkHub captures the texture of student work as it happens. Every revision, every pause, every breakthrough. When Aidan, our AI tutor, helps a student think through a problem, the teacher can see exactly how the conversation went. They can review their classroom not as a grid of grades, but as a record of how thinking unfolded.

The point is not to add another dashboard to a teacher's day. It is to give the teacher back the missing information that drains them when they have to operate without it. We have spent years asking how AI can help students. The Marshall and Pressley study suggests the question we should also be asking is quieter and just as important.

If the right kind of visibility can lower a teacher's anxiety and let them feel like a teacher again, what are we still building that does the opposite?