Norman Eng Just Gave Faculty the Right Questions About AI. Here Is What Comes After Yes.
Norman Eng published something in Faculty Focus this morning that every educator should read before their next syllabus revision.
The Faculty Focus piece is titled Should We Integrate AI into Our Teaching? Evidence-Based Guidelines for Deciding When AI Belongs, and it does what almost no AI-in-education piece has done in the last two years. It treats the question as a decision, not a foregone conclusion.
Eng is a lecturer at Brooklyn College’s School of Education and the founder of EducationXDesign. He has spent his career thinking about how durable learning actually happens, and he is one of the first respected voices to look at the AI-in-classroom rush and ask the question that should have been asked first.
Does the evidence support this?
The Evidence Problem Eng Names
Eng walks through the research that should be putting the brakes on AI deployment in higher education. The findings are converging, and they do not say what the marketing decks say.
- A Swiss study found a negative correlation between AI tool use and critical thinking. The more students offloaded cognitive work to AI, the weaker their critical thinking became (Gerlich, 2025).
- Wharton researchers documented what they call cognitive surrender, a wholesale transfer of agency to the machine, not just deliberate offloading. Across multiple experiments, participants accepted AI-generated outputs with little or no scrutiny (Shaw and Nave, 2026).
- A 2025 meta-analysis of eighteen GenAI studies confirmed that over-reliance on AI tools undermines higher-order thinking, including critical analysis and problem solving (Qu et al., 2025).
- A 2025 randomized controlled trial, the gold standard of educational research, found that students who used ChatGPT as a study aid retained significantly less knowledge 45 days after instruction than students who studied without it (Barcaui, 2025). The short-term gains masked long-term loss.
And per an analysis of 65 R1 institutions, 63 percent of them are actively encouraging GenAI adoption anyway (McDonald et al., 2025). Adoption is racing ahead of evidence. That is the gap Eng is naming.
Eng’s Four Questions
Eng builds his framework around four conditions known to produce durable learning: a strong content knowledge base, deep processing and productive struggle, independent critical thinking, and meaningful human interaction.
Before integrating AI into any lecture, activity, or assignment, he says, work through these four questions:
Will this AI tool help students use, recall, and demonstrate understanding of core disciplinary content?
Higher-order thinking is built on a foundation of domain knowledge. If an AI tool actively engages students with foundational content through retrieval, feedback, or elaborative questioning, it may earn a place. If it lets students bypass the content, it is almost certainly counterproductive.
Will this AI tool require students to apply their learning to a new context?
Transfer is one of the most reliable indicators of genuine understanding. AI that scaffolds transfer while preserving cognitive effort can be valuable. AI that performs the transfer for the student short-circuits learning.
Will this AI tool support, not replace, independent evidence-based reasoning?
Critical thinking requires students to make judgments and defend them. Eng’s test is brutally simple: after the task, can the student articulate, in their own words, why they made the decisions they made?
Will this AI integration preserve meaningful human interaction?
Peer feedback, collaborative problem-solving, and student-to-instructor dialogue do more than support academic learning. They build the habits that define educated citizens. AI integration that quietly replaces these interactions sacrifices more than it gains.
Eng’s default is what he calls offline pedagogy: design instruction around the conditions known to produce durable learning, and integrate AI only when it can be shown to support those conditions rather than substitute for them. The burden of proof, he says, belongs to the technology, not the faculty member who questions it.
He is exactly right.
Like what you are reading? Get insights like this delivered daily.
Free community. Daily Seed videos. Working frameworks instead of vague advice.
The Integration Gap Eng’s Framework Does Not Address
Here is the part the four questions do not solve, and the part that matters most for the educator who reads Eng’s piece on Monday morning and has to make decisions on Monday afternoon.
The questions tell you whether AI belongs in a given lesson. They do not tell you how to structure the AI use across a semester so that the yes answers compound into actual skill, instead of decaying into the same cognitive surrender Eng’s sources are warning about.
A decision framework without an integration model is half a tool.
You can ask all four questions, get four honest yes answers, and still end up with students who quietly offloaded the work, because nobody designed the arc of how AI use should evolve as they grow. That is the gap. And it is exactly the gap SeedStacking was built to close.
SeedStacking as the Integration Layer
SeedStacking is the Harvest Kernel methodology for building AI fluency through small, structured daily wins that compound over time. It has four progressive phases, and they map almost exactly onto Eng’s four learning conditions.
Matches Condition 1: build content knowledge
At the Seed phase, AI is in the smallest possible role. You are not asking AI to do the work. You are using AI to reinforce content knowledge through retrieval practice, targeted feedback, or one elaborative question. The student does the thinking. AI prompts them to do more thinking. Five-minute interactions, daily, focused on one specific concept.
Matches Condition 2: protect productive struggle
At the Sprout phase, students attempt application before consulting AI. The AI check happens as a comparison after they have already wrestled with the problem. The productive struggle is preserved because the AI never sees the work until the student has already done it.
Matches Condition 3: defend independent reasoning
At the Grow phase, AI becomes a critique partner, not a producer. Students draft, reason, and defend. Then they ask AI to push back. The judgment stays with the student. AI provides counter-arguments the student must evaluate, name, and either accept or refute in writing.
Matches Condition 4: preserve human interaction
At the Harvest phase, students teach. They produce a shareable artifact, a lesson, a write-up, a presentation, that demonstrates their own thinking. They present to peers. Peers question. Discussion happens. AI is referenced as one tool among many, never as the source of the work. The human interaction Eng wants protected is the literal endpoint of every SeedStacking cycle.
This is the integration model that takes Eng’s four yes answers and turns them into structure instead of risk.
The Honest Test
Combine Eng’s four questions with the SeedStacking phases and you get one honest test for any AI-in-classroom proposal:
If you cannot explain how AI use at each phase builds to the next phase, you are not integrating AI. You are deploying it.
And deployment without integration is the exact pattern that produces the cognitive surrender Wharton documented. The exact pattern that turns Khanmigo into a non-event for the students who needed it most. The exact pattern Eng is warning faculty to refuse.
Eng is right that the burden of proof belongs to the technology. SeedStacking is the framework that lets the technology actually earn its place.
Read Eng. Then ask one more question.
The four questions are the cleanest decision filter on AI in higher education published in 2026 so far. They tell you whether AI belongs in a given lesson.
SeedStacking tells you how to structure the yes answers so they compound into skill instead of decaying into surrender. Read his piece, then ask one more question of your course design: what comes after yes?
The decision framework tells you when. SeedStacking tells you how.
The free Harvest Kernel community is where educators, professionals, and lifelong learners work through real AI-integration decisions together. Bring a lesson plan, a syllabus question, or a workflow you are trying to design. We will work through it with you.
