An assistant for teaching
The idea of the augmented teacher predates generative artificial intelligence. For years, digital platforms, learning management systems, and assessment tools have helped teachers organize parts of their work. But today’s new AI tools are no longer limited to storing information, automating a task, or displaying results. They can generate materials, suggest activities, adapt texts to different levels, create questions, summarize information, or recommend strategies for a specific classroom challenge.
This shift changes the teacher’s relationship with technology. A traditional platform requires users to learn what functions it offers and then use them for a particular task. With generative AI, the starting point can simply be a need: “I need to explain this in a different way,” “make this activity easier,” “suggest three exercises for practicing fractions,” or “how can I organize this lesson for students with different reading levels?”
Teachers are already incorporating these possibilities into their work. According to TALIS 2024 (the OECD survey on the state of the teaching profession), approximately one in three teachers uses artificial intelligence in their professional work across participating education systems. Among them, 68% use it to learn about or summarize a topic, and 64% to generate lesson plans or activities. Its use is less common for tasks such as assessment or analyzing student performance, but a pattern is beginning to emerge: AI is making its first inroads where it can facilitate the preparation and organization of teaching.
Some educational tools are taking this logic a step further. Rather than expecting teachers to find the right application for each need on their own, they are beginning to bring different forms of assistance together within a single system. CENTA, for example, offers lesson-plan and resource generation alongside professional development content. Other platforms use AI to support assessment, interpret information about learning, or adapt activities.
The purpose of these tools is to handle part of teachers’ workload with the help of an assistant, giving them more time for other tasks. But to understand what that time saving might actually mean, we first need to look at how teachers currently spend their working hours.
The work that happens before and after class
Teaching takes up only part of a teacher’s working day. According to TALIS 2024, full-time teachers spend an average of around 43% of their working time on teaching itself. Lesson preparation accounts for another 14%; marking and assessment, 9%; and administrative tasks, approximately 6%. More than half of their working day, therefore, is spent on activities other than directly teaching students.
Much of this work is also a significant source of pressure. Thirty-five percent of teachers report experiencing quite a lot or a lot of stress because of excessive lesson preparation, while 52% point to administrative work. This is precisely where generative AI offers its most immediate benefits: producing a first draft of an activity, creating questions, adapting a text, or generating examples takes seconds.
But it is one thing for a task to be completed more quickly and another for those savings to make a meaningful difference to a teacher’s working day. A randomized trial conducted in England by the Education Endowment Foundation and the National Foundation for Educational Research offers an initial indication. The study followed 259 science teachers in 68 secondary schools. One group used ChatGPT, together with guidance on its use, to prepare lessons and materials; the other worked without generative AI. Teachers in the first group spent an average of 56.2 minutes per week preparing the lessons included in the study, compared with 81.5 minutes in the comparison group: 25.3 minutes less, a reduction of 31%.
An independent panel also reviewed a sample of the materials produced and found no appreciable differences in quality, although the researchers caution that this part of the analysis was based on a limited sample.
The finding is also interesting because of how teachers used the tool. ChatGPT was used mainly to generate questions and quizzes, find ideas for activities, or modify existing materials. AI was not preparing lessons instead of the teacher. It was taking care of small pieces of their work.
The accumulation of these small forms of assistance may ultimately prove more significant than the most spectacular applications of the technology. For now, the augmented teacher begins with having help to do certain things faster. The next step is having help to decide how to do them.
El profesor aumentado empieza, de momento, por disponer de ayuda para hacer algunas cosas más rápido. Lo siguiente es disponer de ayuda para decidir cómo hacerlas.
From continuous professional development to continuous support
Teachers’ continuous professional development has traditionally been organized around courses, workshops, and training sessions. Teachers learn a methodology, discover new resources, or receive guidance and then return to the classroom, where they have to apply that knowledge to a particular group and deal with situations that no course can fully anticipate.
The issue is not only what training teachers receive, but when they can access it. TALIS 2024 shows that 63% of teachers cite lack of time as a barrier to participating in professional development activities. Nearly 60% mention conflicts with their work schedules, while 46% point to cost.
Research has long examined an alternative: ongoing support. A meta-analysis of 60 studies using causal designs on teacher coaching found positive effects of this type of support on teaching practices and also, though smaller, on student achievement. However, the results diminished as programs grew and had to serve larger numbers of teachers. What works with a coach who can closely support a small number of teachers is more difficult to reproduce for thousands.
Artificial intelligence presents a different possibility: providing some form of professional support when the need arises and doing so at a scale that is difficult to achieve through face-to-face support.
That is what aprendIA is attempting to do. Developed by the International Rescue Committee for teachers working in vulnerable and crisis-affected settings, the tool operates through messaging platforms such as WhatsApp and combines short professional development courses with responses to specific challenges involving classroom management, literacy, mathematics, or social-emotional learning.
The questions asked by teachers involved in its deployment in northeastern Nigeria help illustrate the difference. They were not only asking broad pedagogical questions. They wanted to know how to manage a large class, regain students’ attention after recess, teach reading, or find another way to explain subtraction. Queries clustered during the school day and in the afternoon, when teachers were preparing their lessons.
The first published results are still preliminary. By the end of 2025, 642 teachers had generated more than 124,000 messages across around 10,000 sessions. In a comparison of 224 participants before and after the program, teachers reported greater confidence in adapting lessons to different needs, maintaining students’ attention, and managing large groups. The project has yet to demonstrate whether this use produces observable changes in teaching practice and, above all, in learning.
The interest of the model lies precisely here. A teacher can complete a short course on classroom management and then turn to the same system when they need to apply what they have learned to a particular situation. Training is no longer something that happens only before teaching; it can accompany practice.
This is particularly relevant in contexts where education systems lack enough trainers, mentors, or specialists to provide ongoing support to every teacher. A phone and a messaging application reduce infrastructure requirements and make it possible to bring some form of support to places where a coach would be difficult to provide.
That does not turn an AI into a mentor. A professional can observe a class, understand the school, interpret relationships, identify difficulties the teacher has not articulated, and build a relationship of trust. AI responds to the information it receives. Its value may lie elsewhere: providing an initial source of help when the immediate alternative is facing the problem without support.
When the assistant also gives advice
There is an important difference between asking an AI to produce ten questions about a text and asking it how to work with a student who is struggling to learn to read. As technology moves from producing materials to recommending pedagogical decisions, the demands we place on it must increase as well.
A system can produce a flawlessly written activity while suggesting an inappropriate strategy. It can provide incorrect information, reproduce biases, or recommend practices that disregard students’ curriculum, language, or cultural context. There is also a risk of homogenizing teaching if many teachers turn to the same models for activities, explanations, and assessments.
Teachers themselves recognize some of these problems. TALIS 2024 shows that seven in ten believe AI can make it easier for students to present someone else’s work as their own, while around four in ten believe it can amplify biases, reinforce errors, or compromise data privacy and security. At the same time, three in four say they lack the knowledge or skills needed to teach using artificial intelligence.
The augmented teacher therefore needs more than tools capable of producing good answers. Teachers need the judgment to decide when to use them, check what they generate, and recognize when a recommendation should be rejected.
UNESCO places precisely this capacity at the heart of its AI Competency Framework for Teachers. Its 15 competencies span five areas: 1) a human-centered mindset, 2) ethics of AI, 3) AI foundations and applications, 4) AI pedagogy, and 5) AI for professional development. The aim is not simply to learn how to write better prompts, but to understand what the tool can do while retaining responsibility for educational decisions.
The same requirement applies to those designing these assistants. Before deploying aprendIA in Nigeria, for example, the IRC subjected the system to tests designed to identify incorrect responses, tone problems, cultural biases, and security vulnerabilities. The tool also uses pedagogical content that has been selected in advance and adapted to the context.
When AI enters the staff room, therefore, more matters than the quality of the model behind it. The sources it learns from, the curriculum it knows, the languages it handles, the data it can access, and the limits placed on its recommendations all matter.
Human agency at the center
The expression “augmented teacher” may bring to mind a teacher capable of producing more: more materials, more exercises, more assessments, and more activities in less time. But reducing the impact of artificial intelligence to productivity would miss much of what is at stake. What matters here is what teachers do with the time and capacity they recover with the help of artificial intelligence.
Noticing why a student has fallen behind. Talking to someone who has stopped participating. Changing an explanation when it becomes clear that the class does not understand it. Recognizing that a carefully prepared activity is not working. Listening, asking questions, and improvising are all part of a job that depends on knowing students and understanding what is happening in that particular moment.
An AI can generate twenty ways to explain a fraction. The teacher still has to decide which one the child in front of them needs.
UNESCO already speaks of a teacher-AI-student relationship and places human agency at its center: teachers’ ability to retain control over decisions and use technology in accordance with pedagogical objectives defined by people.
We still know little about how that relationship will evolve. The evidence on time savings is promising but limited. Experiences such as aprendIA are still being evaluated, and important questions remain about quality, dependence, privacy, bias, and the effects on teachers’ own professional competence.
But AI has already begun to occupy a place in lesson preparation, assessment, professional development, and the small decisions surrounding every hour of teaching. The augmented teacher is beginning to exist. So now we need to start thinking about which parts of their work we want to augment, and for what purpose.
