Point of view

The biggest barrier to AI in schools is skills, not technology

OECD's TALIS 2024 finds that 27% of Finnish lower secondary teachers have used AI at work. The reason the rest have not is rarely the equipment.

Jaakko Rannila, founder, Bruukki
The biggest barrier to AI in schools is skills, not technology

In public debate, AI is already everywhere in schools. An international teacher survey says otherwise. In the OECD’s TALIS 2024 study, 27% of Finnish lower secondary teachers reported having used AI in their work. The OECD average is 36%.

The adoption rate is not the interesting part. What matters more is what teachers use AI for and what holds the rest back. We went through the figures from the perspective of making learning materials.

Use concentrates on preparation

Among Finnish teachers who use AI, the most common uses are learning about and summarising a topic (48%), producing lesson plans (40%) and having students practise new skills in real-life situations (36%). The rarest are assessing and marking work (23%), reviewing performance data (14%) and writing student feedback or messages to guardians (13%).

Teachers, in other words, draw the line themselves, and they draw it where Finland’s national AI recommendations draw it: AI may help with preparation, the assessment decision belongs to a human. The same emphasis shows in what teachers see as the benefits. Across OECD countries, around half of teachers think AI assists in creating or improving lesson plans, and 40% agree it helps them support an individual student. Country differences are wide: on lesson plans the share runs from 18% in France to 91% in Viet Nam.

Bruukki is built for exactly that part of the work. You make materials, exercises and difficulty versions in one place, and every suggestion comes to you for review. No suggestion reaches a student before you have gone through it. AI suggests, the teacher decides.

The barrier is skills, not hardware

The clearest finding concerns those who do not use AI. Of the Finnish teachers who had not used AI in their teaching over the past year, 81% said they lack the necessary knowledge or skills, against an OECD average of 75%. Far fewer were held back by equipment or networks: 24% said their school lacks the infrastructure AI use requires, where the OECD average is 37%.

So infrastructure blocks teachers in Finland less often than in OECD countries on average, and use is still below average. The skills gap also shows in what teachers ask for themselves: in the TALIS data, the top professional learning need Finnish teachers named was using AI in teaching and learning (23%). The OECD’s Digital Education Outlook 2026 adds one more finding. Of the teachers who report lacking the skills, about half also believe AI should not be used in teaching. Missing competence and a negative stance tend to travel together.

Two things follow. First, the tool has to fit the teacher’s task as it stands. A tool that requires learning to write prompts filters out users from precisely the group that should be getting started. In Bruukki the starting information comes from the curriculum and from your own material, so there is no empty prompt box to fill.

Second, no tool replaces training. Competence is also an obligation: the EU AI Act requires staff to have sufficient AI literacy. This is worth handling systematically rather than leaving it to each teacher’s own experiments.

The teacher’s tool does not change when the model does

The skills question has a side that is easy to miss. Language models move fast, and this year’s best model is not necessarily the best one next year.

In a general-purpose AI tool the model does get updated underneath without the interface changing. What an institution does not get is a say in which model runs or when it changes, and the choice stays inside one provider’s range of models. If an institution’s own assessment points to another provider’s model, on quality, price, data protection or its own supplier policy, switching model means switching tools. The teacher’s interface and ways of working then change with it.

In Bruukki, which language model is used is a workspace setting agreed with the institution, so the service is not tied to one model. A teacher does not have to follow that debate. When the model changes, the views, the materials and the ways of working stay the same, and nobody has to learn a new tool. We covered this in more detail in what schools should pay for.

For the skills gap this is a large difference. A teacher does not have to learn how AI works or how to steer it to get a usable result. The tool does that work, and it stays in place even when the model underneath it changes.

The worry about cheating and errors is justified

Teachers’ reservations appear in the same data. Seven in ten teachers across OECD countries think AI lets students pass off others’ work as their own. Around four in ten think AI may amplify biases and reinforce student misconceptions, or compromise data privacy.

The time saving is not a given either. The OECD review points to interviews with teachers in Sweden and Australia, who report investing significant work in reviewing, repairing and reworking AI-generated output. The promised time saving disappears if every suggestion has to be rewritten.

The design of the tool matters here. The risk of wrong suggestions and rewriting drops when the AI leans on a source instead of guessing. In Bruukki a material can be based on your own files, on the curriculum or on another source you have saved. The cheating question is not one any tool answers for you. It is answered through assignments that make the learning process visible, and AI is good help in writing those.

In a Nordic survey commissioned by Kopiosto, between 15% and 33% of Finnish teachers, depending on school level, said they had uploaded someone else’s material into an AI tool in the past month. The share varies, but the pattern is clear. Under time pressure, material ends up in the tool without anyone finishing the thought about rights.

This is worth separating from data protection, even though both apply at the same moment. Copyright decides whether the material may be entered into the tool. Data protection decides what happens to it afterwards. In Bruukki the second part is written down in advance: data and materials are stored in the EU (AWS Ireland), and Bruukki works under Zero Data Retention terms, so Bruukki leaves none of your material with the language model provider and does not let it be used to train language models. The first part stays with the teacher and the institution: your own material, open collections and material the institution holds the rights to.

In higher education, use is already widespread

Education level moves these numbers more than any other single factor. The OECD review highlights a French survey of 30,000 higher education students, teachers and staff. In it, 80% of higher education teachers reported having used generative AI in 2025, most often to prepare and write their course (49%).

In higher education the work is well under way, and the question has moved from use to practice. In Bruukki, a course’s learning outcomes or an institution’s own curriculum can be taken as the basis for the material, so the same way of working serves a lecture course as well as a school subject. We wrote more from the higher education angle in what FINEEC expects from higher education teaching.

What schools should do

Three tasks follow from the figures. Handle staff AI literacy systematically. Choose and approve tools centrally, so teachers do not have to assess data protection alone. Agree what AI is used for and what it is not, and tell students the same thing.

Try it on your own material

The best way to judge a tool is to try it on real work. Take an old handout or a set of course objectives, bring them into Bruukki and see how close the first suggestion gets. Try it for free.

For institutions we offer a free trial period and provide data protection documentation to support the evaluation. A demo takes about 30 minutes. Get in touch at bruukki.com/en/contact/.

Read also: what schools should pay for and support for the whole teaching week.

Sources

Bruukki's AI was used to help write this blog post.

Try it with your own material

The AI suggests, you decide.