In one Moscow school, a computer science teacher took just a week to assemble an AI agent from ready-made modules to analyze students' algebra mistakes and provide personalized assignments. Notably, he did not write a single line of code.
The method is based on the Activity Theory of Aleksei N. Leontiev and Yrjö Engeström. Writing on arXiv (abs/2605.12934), researchers described how six components—subject, object, tools, community, rules, and division of labor—help teachers turn the abstract challenge of creating an agent into a clear sequence of actions.
First, the teacher defines the object, such as reducing the workload of grading tests. They then select the tools, utilizing ready-made platforms like Teachable Machine or LangChain. Next, they establish the rules: student data stays at the school, and the agent cannot make decisions without teacher confirmation. A community of colleagues discusses potential use cases, while the division of labor assigns roles for data management, ethical oversight, and classroom integration. This framework enables teachers to see a manageable system rather than an opaque "black box."
Preliminary data from a pilot study across three Russian regions shows that 78% of participants were able to independently deploy an agent for their subject area after a 24-hour course. However, the sample size remains small, there was no control group, and the long-term impact on student achievement has not yet been measured. Critics also highlight the risk that teachers without a deep understanding of algorithms might place blind trust in an agent’s outputs.
This approach exposes the tension between the increasing accessibility of AI tools and the persistent lack of time teachers have to master them. When building an agent becomes a routine part of their work rather than a standalone project, the barrier to entry is lowered, but the school’s dependence on external platforms and their privacy policies inevitably grows.
The primary question is no longer whether teachers can create AI agents, but rather what rules and communities they can build around these tools to maintain control over the educational process.
