Agentic AI can preserve thinking when students have to design the work

Claim

Agentic AI can preserve or deepen cognitive work when users must define the purpose, audience, structure, constraints, and success criteria.

Supported by the source articles as an AI-in-education claim.

Helpful AI is the problem, we’re the solution supports this claim through its discussion of high relevance for productive friction, AI simulation design, formative assessment, transcript-based evidence of thinking, and alternatives to AI detection.
The Box and the Module distinguishes one-shot AI generation from AI-assisted design work where teachers must specify pedagogical goals and constraints.
The Box and the Module supports this claim through its discussion of AI literacy, assessment, implementation, or learning design in context.
We could be building a better world with our students. Why are we disempowering them instead? supports this claim through its examples of students building classifiers, auditing outputs, revising definitions, and debating project goals instead of passively accepting AI-generated answers.
In less than five hours, I wrote a textbook and course handbook with AI … and both are good supports this claim by showing that AI-assisted drafting preserved substantial human thinking only when Van Slyke had to specify audience, architecture, standards, skill goals, and productive-friction constraints before and after generation.
The Writing Revolution 2.0, Chapter 1 provides a non-AI analogue for this claim: specific writing constraints such as “because, but, and so” can preserve cognitive work by requiring students to retrieve, contrast, explain, and justify ideas rather than answer passively or open-endedly.

Educators should distinguish AI uses that bypass thinking from AI uses that require substantial design judgment and iterative evaluation.