Your child’s next AI tutor might be so helpful that it actually stops them from learning. New research warns that as AI evolves into proactive "agents" that take initiative, they risk smoothing away the mental struggle required to build long-term mastery.
AI is shifting from passive chatbots to proactive assistants, but if these "agents" automate too much of the process, your child will stop developing critical thinking skills. Effective learning requires a specific amount of "intentional friction" that most current AI tools are designed to eliminate.
Efficiency is the enemy of retention. We often judge educational tools by how quickly they help a child finish a task, but speed is not a proxy for learning. If a tool makes a difficult subject feel too easy, your child is likely just outsourcing their brain to the software.
This creates a hidden "learning debt." When a child uses a frictionless tool to breeze through algebra or essay writing, they aren't building the neural pathways necessary to solve those problems independently. Choosing "seamless" apps today may result in significant knowledge gaps tomorrow when the AI isn't there to hold their hand during a test or a real-world challenge.
Researchers are concerned that we are entering an era of "agentic" AI—systems that don't just wait for a prompt, but take the initiative to solve problems and complete goals. While this shift is a massive win for office productivity, it creates a fundamental conflict in the classroom.
The authors analyzed core learning principles against these new AI capabilities to see where they clash. They found that the very features tech companies use to market AI—proactivity, speed, and ease of use—are the exact features that can undermine "cognitive effort," which is the mental work required to actually store and use new information.
There is a direct tension between AI automation and human learning. To truly learn, a student must wrestle with a concept, a process known as "desirable difficulty."
- Automation kills struggle: AI agents are being built to minimize friction, which is the opposite of what a developing brain needs to grow.
- Intentional friction is necessary: Productive learning requires a specific level of difficulty. If the AI "smooths" the path too much, the student never engages deeply with the material.
- The need for dynamic scaffolding: Effective AI should provide "just enough" help to keep a student moving (scaffolding) and then pull back support as the student gains competence.
- The "black box" problem: When an agentic AI takes over a task, it often hides the steps it took to get to the answer, preventing the student from seeing the logic required to reach the conclusion.
The tech industry’s obsession with "seamless" user experiences is a pedagogical nightmare. For a consumer product, "no friction" is the gold standard of design. For an educational tool, "no friction" is a failure.
We are currently using tools designed for corporate efficiency—tools meant to save time for busy adults—to teach children. This prioritizes the output (the finished essay or solved equation) over the process (the child’s understanding). If your child is using a tool that "anticipates" what they want to say or "suggests" the next three steps in a math problem, the AI is the one doing the learning, not the student.
This paper is a theoretical review and design framework, not a controlled experiment with real-world student data. It outlines how AI should be built based on learning science, but it hasn't yet proven exactly how much damage current "un-agentic" tools are doing. Additionally, the paper is a preprint and has not yet undergone formal peer review. It is a roadmap for future design rather than a report on existing classroom outcomes.
- If you are choosing a new tutoring app for your child, look for features that simulate a Socratic tutor—asking "What do you think the next step is?"—rather than an assistant that simply provides the answer or finishes the sentence.
- If your child is using AI for homework, sit with them to ensure they are using it for "metacognition." Ask them to have the AI explain the logic behind a step rather than just generating the result, and then have the child explain that logic back to you.
- If a tool makes a difficult subject feel suspiciously easy, have your child try a similar problem on paper without the app. This "analog check" will reveal if they have actually achieved mastery or if the software was just doing the heavy lifting.
- If you notice an AI tool is "taking initiative," such as suggesting whole paragraphs or solving equations before being prompted, disable those proactive features in the settings to force the child back into the driver's seat.
- If your child is stuck on a prompt, encourage them to use the AI to brainstorm "three different ways to start" rather than asking it to "write the introduction." This keeps the final decision-making and creative effort with the student.
Automation is the enemy of education. If a tool is doing the thinking for your child, it isn’t a tutor; it’s a ghostwriter. Prioritize AI tools that act as temporary "scaffolding" to be removed later, not permanent crutches that make your child faster at the expense of making them smarter.
Steve Woollaston, Brendan Flanagan, Isanka Wijerathne et al. (2026). Agentic AI and Pedagogical Best Practice: The Tension Between Automation and Learning. arXiv (preprint). — arxiv.org