The most important question about artificial intelligence in schools is not whether AI can do some of the things teachers currently do. Of course it can generate lesson ideas, draft questions, adjust text complexity, create rubrics, summarize information, produce family communications, analyze patterns, generate examples, and perform an expanding range of tasks that once required significantly more human time. The more consequential question is what education systems should ask AI to do and what parts of teaching they should deliberately protect from automation.
If we design artificial intelligence around replacing the professional judgment, relationships, creativity, moral responsibility, and human connection of teachers, we will use one of the most powerful technologies of our time to misunderstand one of society's most human professions. If we design AI around protecting teachers from unnecessary burden and increasing the amount of time they can spend on the work only human beings can do well, the technology becomes far more interesting. My working thesis is therefore simple: AI should not replace teachers; it should protect their time, attention, relationships, judgment, and capacity to think deeply about children.
Protecting teachers also means protecting the intellectual energy necessary to plan strong instruction and the emotional bandwidth required to return to students fully present. It means reducing the extraordinary volume of low-value administrative and repetitive work that education systems have gradually allowed to accumulate around the profession. AI should be judged by whether it gives meaningful professional capacity back rather than by how many functions a vendor can say the technology performs.
Teachers Have a Time Problem Before They Have an AI Problem
Education cannot have a serious conversation about artificial intelligence without having an equally serious conversation about teacher workload. Across OECD education systems, roughly half of teachers report excessive administrative work as a source of job-related stress, and that burden matters because every hour consumed by unnecessary administration is an hour of professional attention that cannot simultaneously be spent studying student work, planning instruction, communicating meaningfully with families, collaborating with colleagues, or recovering enough to return to children fully present. The challenge is not simply that teachers are busy; it is that too much of their limited time is often allocated to work that does not require their highest professional expertise.
This is the lens through which districts should evaluate AI because the return on investment should not begin with how many employees a technology allows the organization to eliminate. It should begin with how much low-value work can be removed from teachers while maintaining or improving the quality of the work that remains. That design principle immediately changes the conversation from substitution to professional leverage.
AI becomes useful when it gives a teacher time back from drafting routine communications, helps generate a first version of differentiated practice questions, organizes information that would otherwise require manual sorting, or creates a starting point that a skilled educator can improve. The human remains responsible for judgment, accuracy, context, and the final educational decision. Technology is at its best when it accelerates work without displacing the responsibility attached to the work.
Emerging evidence suggests there may be real workload potential here. In a randomized Teacher Choices trial conducted by the Education Endowment Foundation and evaluated by NFER, 259 teachers across 68 English secondary schools participated in a study of ChatGPT-assisted lesson and resource preparation for Year 7 and 8 science. Teachers using ChatGPT alongside implementation guidance spent substantially less time on the relevant lesson and resource planning than the comparison group, while an expert panel did not identify an apparent decline in resource quality.
That finding is promising, but it should not be inflated beyond what the study actually tested. It involved a particular group of science teachers, a particular planning context, and a particular form of guided AI use, so it does not establish that AI will reduce every teacher's workload by the same amount or that every AI-generated resource is educationally sound. What it does demonstrate is more important than hype: under some conditions, teachers can use generative AI to reduce time spent on legitimate professional tasks without an obvious loss of quality.
The Teacher Should Remain the Professional of Record
The distinction between assistance and replacement has to become explicit. Artificial intelligence can generate an instructional task, but it does not know the children in the room in the same way their teacher does, and it can analyze text, identify patterns, and produce plausible recommendations without carrying professional or moral responsibility for what happens to a child. Schools need a clear human accountability rule so that efficiency never becomes an excuse for shifting consequential judgment onto software.
UNESCO's guidance on generative AI adopts a human-centered approach that emphasizes human agency, ethical use, inclusion, privacy, and the need for technologies to benefit and empower teachers and learners rather than substitute technology for human capacity. Its teacher competency framework similarly emphasizes human-centered mindsets, AI ethics, pedagogy, professional learning, and protection of teachers' rights and agency. Those principles should become practical design rules inside school systems rather than remaining abstract policy language.
The teacher should remain the professional of record. AI can recommend while the teacher decides, AI can draft while the teacher reviews, and AI can identify a possible pattern while the teacher interprets that pattern within knowledge of the student, curriculum, context, and classroom. The distinction becomes particularly important when decisions carry consequences involving discipline, disability, family referrals, teacher evaluation, or future student opportunities.
Those decisions involve context, rights, judgment, evidence, and human accountability that cannot simply be delegated to software. The U.S. Department of Education's AI toolkit likewise emphasizes safe, ethical, and equitable integration, including privacy, security, civil rights, transparency, digital equity, and meaningful understanding of AI-enabled applications. These are not peripheral compliance issues because educational technology operates around unusually sensitive information about children and adults.
Use AI First Where Teachers Are Doing Work Machines Can Help With
The first frontier should be work that consumes teacher time without requiring the deepest forms of human relationship or professional judgment. Drafting routine communications is an obvious example because AI can create a first version that a teacher then verifies and personalizes, while generating multiple practice examples or alternate versions of an assignment may be another. The strongest use cases remove friction from the work without removing the professional thinking that gives the work educational value.
Lesson preparation provides another promising use because the teacher can remain the instructional designer while using AI as a brainstorming and production assistant. A skilled teacher might ask for several examples of a concept, possible misconceptions students may demonstrate, draft discussion questions, additional practice items, or alternate explanations. The professional value lies in the teacher's ability to decide which output is accurate, rigorous, appropriate, and connected to what students actually need.
Teachers are already experimenting with these uses. RAND's nationally representative survey work found that during the 2023–24 school year, approximately one quarter of surveyed teachers reported using AI tools for instructional planning or teaching, while principals reported even higher use in their own work. Teachers described using AI for activities, assignments, lesson ideas, teaching resources, assessments, rubrics, differentiation, and other instructional tasks, although adoption varied substantially across subjects and school contexts.
The variation should concern education leaders as much as the adoption itself because RAND found lower reported AI use among teachers and principals in higher-poverty schools than in lower-poverty schools, with principals in high-poverty settings also less likely to report providing guidance. If AI eventually becomes a meaningful professional capacity multiplier, unequal access to tools, training, infrastructure, and guidance could become another way existing organizational inequalities widen. Equity therefore belongs inside AI implementation from the beginning rather than being added after adoption has already become uneven.
Districts cannot simply purchase licenses and call the work complete. Teachers need guidance on what tools are approved, what data can be entered, how outputs should be checked, where AI use must be disclosed, what kinds of decisions require direct human judgment, and how intellectual property, privacy, bias, and inaccurate output should be handled. Professional development should include actual practice with real instructional tasks rather than a generic presentation about what artificial intelligence is.
AI Should Return Time to Relationships
The highest-value use of AI may ultimately be measured by something the technology never does itself. If a teacher spends less time completing repetitive administrative work, the real question is whether the teacher has more time to talk with students, and if planning becomes more efficient, the question is whether the teacher has more cognitive capacity to analyze the needs of a student who is struggling. Efficiency should therefore be connected deliberately to human reinvestment.
Time saved by AI should be reinvested in the parts of education where humans create distinctive value: relationships, feedback, judgment, mentoring, motivation, explanation, diagnosis, creative lesson design, collaboration, family partnership, and the countless interactions through which a child comes to believe that somebody in the building actually knows them. UNESCO has continued to emphasize that teachers remain fundamental because technologies do not replicate the human understanding of learners' emotions, social relationships, and context. AI can participate in the instructional ecosystem without becoming the heart of it.
That principle should affect procurement decisions. A district should be more interested in an AI product that gives teachers meaningful time back than one that simply creates another dashboard teachers are required to monitor. A technology that saves 30 minutes but produces 45 minutes of additional documentation is not innovation; it is administrative substitution dressed up as progress.
Do Not Turn AI Into a New Surveillance System
There is a darker version of AI adoption that schools should resist. The same technologies capable of reducing teacher burden can also be used to increase surveillance, generate constant productivity metrics, monitor employee behavior, automate performance judgments, and create more administrative demands than they eliminate. Education systems could easily use AI to make already overmeasured adults feel as though every minute of their professional lives is being scored.
That would be an extraordinary missed opportunity because if AI produces efficiencies and leadership responds by filling every recovered minute with another task, teachers will experience no meaningful benefit. The organization will simply increase output expectations until the time savings disappear. The better contract is explicit: when artificial intelligence eliminates unnecessary work, at least some of that capacity should return to the professional rather than automatically becoming another organizational extraction.
Schools should want teachers who have enough time to think, collaborate, study student work, build relationships, and improve their craft. Protecting teacher capacity is not softness because exhausted professionals make worse decisions and have less room for reflection, creativity, and relationship-building. Capacity is a workforce strategy, and AI should be evaluated partly by whether it strengthens that capacity.
AI Should Improve Access Without Lowering Expectations
Artificial intelligence may also have meaningful potential to help teachers respond to variation among students. A teacher can use AI to draft different explanations of the same concept, adapt the language complexity of a text, generate additional practice, develop examples connected to student interests, translate family-facing materials, or create a preliminary set of supports for multilingual learners and students who need additional access. These uses can expand options for teachers without requiring them to create every variation manually from scratch.
RAND's qualitative reporting found teachers already describing AI-supported differentiation, including adjustments to reading passages, word problems, prompts, assignments, lesson plans, and questions. The existence of these uses is encouraging, but the quality of the resulting differentiation still depends on teacher judgment because simplifying material can accidentally reduce rigor and automated adaptations can introduce errors or distort important content. Access should therefore increase without turning differentiation into lowered expectations.
This is why AI literacy becomes a professional competency rather than an optional technology skill. Teachers need enough understanding to interrogate output, recognize bias, verify information, protect student information, and understand that confident language from a model is not the same thing as verified truth. The professional who uses AI well is not the person who accepts its output fastest but the person who knows what requires scrutiny.
AI Should Strengthen Teachers, Not Deskilling Them
There is another danger that deserves more attention. A technology designed to save teachers time can eventually weaken teacher expertise if educators outsource the thinking through which expertise develops, because writing lesson questions, anticipating misconceptions, analyzing student work, and designing explanations are not merely chores. Sometimes they are the intellectual work through which teachers become better teachers.
Leaders therefore need to distinguish between automating unnecessary work and outsourcing professional thinking. UNESCO's guidance acknowledges this tension because generative tools can accelerate planning and content production while a human-centered approach must preserve agency and the development of human capacity. The objective is to remove friction without removing the cognitive apprenticeship through which professional expertise grows.
A new teacher who asks AI to generate every lesson before learning how to design instruction may save time while losing an important developmental experience, while an expert teacher who uses AI to produce ten additional examples after already determining the instructional architecture may be using the same technology in a completely different way. The question is not simply whether AI did part of the work. The better question is whether AI removed work that did not require the teacher's highest expertise or removed the very thinking through which expertise develops.
Give Teachers Agency in the AI Transition
Teachers should not discover the district's AI strategy through a policy memo written after the technology has already been purchased. The people doing the instructional work should participate in determining where AI helps, where it creates risk, what tasks are worth automating, and what forms of professional judgment should remain protected. Implementation will be stronger when educators are treated as co-designers rather than end users.
This is especially important because adoption is moving faster than many formal systems. OECD's TALIS reporting has shown substantial teacher use of AI alongside a significant need for additional professional development, which means the gap between adoption and expertise should be treated as a leadership priority. Banning the technology entirely does not automatically build responsible judgment, while encouraging unrestricted experimentation creates obvious privacy, quality, equity, and academic integrity problems.
Schools need a middle path built around approved tools, clear boundaries, meaningful training, professional judgment, and continuous evaluation. Teachers should know what they are allowed to do, what they are expected to verify, and what the organization will never outsource to automation. Clear governance can increase responsible use rather than merely restrict it.
Measure AI by What It Gives Back
Education will soon have no shortage of AI products promising transformation. Some will genuinely improve professional work, some will produce modest efficiencies, some will prove ineffective, and others will create problems schools have not yet anticipated. The existence of artificial intelligence is not evidence that every use of artificial intelligence is progress.
Schools therefore need a simple test centered on what the technology gave back. Did it give teachers time, give students stronger feedback, make high-quality instructional resources easier to create, make information more accessible to families, reduce repetitive administrative work, or help educators identify something important sooner? Did it preserve human judgment and privacy while doing so, and did the time saved actually return to higher-value educational work?
If the answer is no, the fact that the technology contains artificial intelligence is irrelevant. Technology should have to earn its place in the school the same way any instructional or operational investment should. The best procurement decision is not the one that sounds most futuristic but the one that measurably improves the conditions under which teachers and students do their work.
The Future Should Be More Human, Not Less
I do not believe the best future of education is one in which children spend their days interacting primarily with machines while fewer adults occupy the building. Children need people who notice when something has changed, interpret a facial expression, understand the history behind a behavior, challenge them, mentor them, and alter what they believe about themselves. Artificial intelligence is not diminished by acknowledging those limits; it becomes more useful when we are clear about what we actually want it to do.
The future I want is one in which a teacher no longer spends Saturday afternoon completing tasks a machine could have handled safely in minutes. I want the teacher to arrive Monday having had time to think, rest, prepare, and return to students with greater intellectual and emotional capacity. I want AI to help a teacher generate a first draft so the teacher can spend more time perfecting the final product, surface patterns so the teacher can investigate them, remove repetitive labor so teachers can perform more meaningful labor, and simplify the machinery surrounding teaching so educators can spend more of their lives actually teaching.
That is not a future without teachers. It is a future that finally recognizes how valuable teachers are and uses technology to strengthen rather than hollow out the profession. AI should not replace teachers; it should protect them, and the strongest measure of success will not be how much artificial intelligence schools use but how much more human the work of teaching becomes because we used it wisely.
About the Author
Candido Brown is the founder and CEO of BetterEd Solutions, a school improvement, transformation, and growth firm helping our nation's underserved schools become happier, healthier, higher-performing places where every child can thrive. His work focuses on school improvement, leadership, organizational design, and the responsible use of technology to increase educator capacity rather than replace professional judgment. BetterEd's approach to AI begins with a human-centered question: what work can technology remove so educators can spend more time on the work students most need humans to do?
Sources and Reporting Notes
This article draws on OECD reporting on teacher workload and AI use, UNESCO's guidance on generative AI in education and its teacher competency framework, the Education Endowment Foundation's Teacher Choices trial on generative AI-supported lesson preparation, RAND research on educator AI adoption, and U.S. Department of Education guidance on safe, ethical, and equitable AI integration. The evidence base on generative AI in K–12 education is still emerging, so the article does not claim universal time savings or guaranteed improvements in instructional quality. The narrower thesis is that AI should be deployed first where it safely reduces low-value burden, preserves human accountability, strengthens teacher capacity, and increases the time available for distinctly human educational work.
Book a 30-minute call