A teacher can ask students to show their thinking. But if the gradebook, pacing guide, policy language, parent communication, and schedule still reward only the polished product, that teacher is working against the system rather than with it. That is the institutional problem underneath generative AI in schools.
The System Around the Assignment
For the first several weeks of this series, I have focused primarily on the learning problem. Generative AI has made a polished student product a much weaker proxy for actual understanding. The question facing educators is no longer simply whether a student used AI, but rather what human cognitive work the student was still required to do.
That line of inquiry led to the vital distinction between productive and exclusionary friction. Some difficulty builds learning; some difficulty blocks access. Schools must determine which friction to preserve, which to redesign, and which to remove entirely. Last week, I wrote about educator sensemaking—how teachers and leaders are already making daily judgments about effort, access, authorship, and evidence of learning.
This week asks the next logical question: What has to be true of the system for those individual judgments to become coherent institutional practice rather than a thousand improvised decisions?
Infrastructural Friction Is the Load-Bearing Layer
In the framework I am developing, three types of friction occur directly within the interaction between the student and the task:
- Noetic friction protects internal thinking (interpreting, retrieving, drafting, revising, synthesizing, and explaining).
- Rhetorical friction protects dialogue (discussion, critique, defense, and revision with real audiences).
- Existential friction protects authorship and accountability (asking whether students can truly stand behind their claims, choices, and evidence).
Infrastructural friction is different. It is not a classroom strategy applied to a specific assignment; it is the macro-level, load-bearing layer beneath the other three. This layer encompasses board policy, assessment expectations, professional learning models, leadership language, procurement, digital infrastructure, privacy rules, grading practices, scheduling, and pacing guides. These systemic conditions dictate whether task-level friction can realistically survive.
A teacher may want students to complete an initial, unassisted attempt before using AI feedback, but if the pacing guide leaves no room for cycles of attempt, feedback, and revision, the practice cannot be sustained. A teacher may want students to defend an AI-assisted essay in a short conference, but if the daily schedule leaves no structural time for conferencing, oral defense becomes a burden the teacher must carry during lunch. Similarly, a team may want to grade the learning process, but if the district-mandated gradebook only accommodates categories for final products, draft histories and revision memos remain optional extras.
AI-era learning design is not merely a classroom problem. It is an institutional design problem.
Policy on Paper Is Not the Same as Guidance in Practice
Schools certainly need AI policies to address privacy, safety, academic integrity, bias, and approved tools. However, policy on paper is not the same as usable guidance. A district may have a comprehensive AI policy on the books, but if teachers cannot translate it into daily classroom decisions, it creates infrastructural opacity. The core issue is not the mere existence of a policy document, but whether educators experience it as usable institutional backing rather than compliance paperwork.
Vague statements like “use AI responsibly,” “AI is allowed,” or “AI is prohibited” are insufficient. Teachers need explicit guidance that connects AI use directly to the learning sequence.
They must be able to answer: When may AI enter the work?
- Before a first attempt as a brainstorming tool?
- After a first draft to generate counterarguments?
- During revision as a language polish?
- As a translation support for a multilingual learner?
Each point in the instructional sequence fundamentally changes the learning question. AI used before an attempt can remove the precise cognitive struggle the task was designed to develop. AI used after an attempt can support meaningful revision. No universal, top-down rule can settle all of these cases, but schools can build systems that make the pedagogical reasoning visible.
AI Policy Is Also Assessment Policy
This systemic reality is most visible in assessment design. If generative AI can produce a flawless final artifact, schools need better evidence of learning than the final product alone. This does not mean every assignment needs to become longer or more complex; it means assessment design must incorporate intentional, verifiable evidence points:
- An initial unassisted attempt
- A verifiable draft history
- A brief oral explanation or source-choice defense
- A revision memo or peer critique record
- A process reflection
These are learning-design moves, not surveillance tactics. They intentionally reconnect product to process.
However, our current digital infrastructure—including traditional Learning Management Systems (LMS) and rigid gradebook structures—is fundamentally built to record static final products. True infrastructural alignment means providing teachers with the institutional permission and the software capability to value the process. If visible thinking is treated as an extra-credit luxury, only a few teachers will utilize it. If it is embedded into the district’s assessment design, teachers have systemic backing, families receive a coherent message, and leaders can defend the practice.
Build Shared Judgment, Not Only Shared Rules
While schools require shared rules for data privacy (such as FERPA, COPPA, or state-level protections), rules alone cannot resolve complex pedagogical dilemmas. Hard questions require shared judgment: Should a student use AI to summarize a source before reading it? Should a student submit an AI-assisted essay if they can defend the claim orally?
Developing this judgment requires a common professional vocabulary. Educators need explicit terms for what they are witnessing: productive friction, exclusionary friction, cognitive bypass, learning sequences, and infrastructural opacity.
This shifts the purpose of professional development. A training session that only teaches educators how to write better prompts treats teachers as passive consumers of tech vendor products. Framework-centric professional learning treats teachers as instructional architects. Educators need structural time to examine concrete student cases, name the specific learning goal being protected, identify the barriers being removed, and determine what evidence of understanding must remain visible.
The Equity Problem
Saying “just trust teachers” sounds respectful, but when institutional support is uneven, individual improvisation quietly advantages already-privileged settings. Educators with more planning time, stronger professional networks, clearer leadership support, better tool access, and more coherent family communication can develop robust AI-era practices. Others are left to figure it out alone.
The same disparity exists across entire school districts. Infrastructure gaps and professional learning resources are not evenly distributed, making infrastructural friction a profound equity issue. We must look critically at who receives clear instructional guidance, high-quality tools, framework-deep professional learning, and leadership backing—and who is left with nothing but compliance language and individual isolation.
The Leadership Question
The leadership task is not to eliminate teacher judgment, but to make that judgment informed, visible, supported, and coherent across a system. This means naming productive struggle as an explicit educational value, treating assessment redesign as a core component of AI governance, and making policy usable in the classroom.
Teachers can preserve productive struggle within the four walls of an individual classroom. But if friction-preserving pedagogy is going to survive at scale, it cannot depend solely on the personal commitment of isolated educators. It must become a structural value embedded within the system around the assignment.
A Question for Leaders
Which structural element is currently the tightest bottleneck for AI-era learning design in your school or district: your pacing guides, your grading and assessment policies, or your current professional development model?
