AI is already part of the classroom.
Students are using it to get through work faster.
Teachers are using it to plan, assess, and personalize instruction.
Schools are starting to see improvements in outputs: grades, completion rates, efficiency. But an important question remains:
Is anyone actually learning?
In my previous piece, I argued that AI isn’t creating a new problem - it’s amplifying an existing one: education systems that are optimized for performance rather than learning.
What are the Threats and Antidotes?
Across recent studies (2024–2026), the threats cluster into three areas:
Cognitive: offloading reduces thinking, memory, and transfer
Behavioral: dependency reduces effort, agency, and persistence
Institutional: assessment collapse weakens signals of learning
The meta-insight is hard to ignore: AI can improve academic performance while degrading learning, if used poorly.
AI should not replace thinking. It should structure and support it.
That means redesigning teaching so AI supports:
planning
feedback
reflection
access
…while the core cognitive work stays with the learner.
This direction is consistent across research from OECD, UNESCO, and recent peer-reviewed studies.
Bottom line: AI should be used to support feedback, planning, coaching, reflection, and access, while the core work of retrieval, reasoning, explanation, judgment, and memory formation remains with the learner.
In practice:
Students attempt first before turning to AI
AI is used as a scaffold, not a substitute for thinking
Students reflect by explaining how their thinking has changed
Outputs are verified against knowledge and reliable sources
Assess authentically to reveals thinking, not just polished outputs
Train teachers to guide effective AI use
To make this actionable, at COGx we translated these risks and responses into a practice guide on what to watch for in the classroom, and what to respond.
How are you seeing these patterns show up in your context?
What’s working and what isn’t?