Borrowing from the Past: Productive Friction in the Age of AI

By: Andy Szeto

Borrowing from the Past: Productive Friction in the Age of AI

Earlier this year, one of my graduate classes gave me a card. Inside, a student wrote: “Thank you for your human intelligence in the age of artificial intelligence.” I kept rereading that line, not only because it was kind, but because of how it arrived. The note was handwritten. The ink smudged a little. It took time.

And that’s what stayed with me: the message, yes, but also the medium. The card did something AI doesn’t do by default. It slowed me down. It made gratitude feel personal, imperfect, and real. In a moment when schools are racing to adopt AI for efficiency, that card reminded me of something we rarely name: learning needs the right kind of friction, the effort, attention, and struggle that make growth meaningful (WestEd, 2025).

AI is great at making things faster. Faster feedback. Faster drafts. Faster translations. Faster “done.” But if school has taught us anything, it’s that faster is not always better. So I keep coming back to a question that feels slightly unpopular in an era of automation: What if the best use of AI is not speed, but productive slowing down?

Not to drag us backward for nostalgia’s sake, but to borrow a few powerful practices from the past and redesign them for the future. In every case, the goal is the same. AI removes friction, so we need to intentionally restore the right kind of friction, the kind that makes thinking visible. 

1) Content knowledge: fewer topics, deeper mastery

When AI can generate answers instantly, the friction of learning content can disappear. Students can look fluent without understanding. That is exactly why content knowledge matters more now, not less. Students cannot reliably verify or fact-check what a tool produces if they do not already have a base of power knowledge: core concepts, key vocabulary, big timelines, and foundational models that organize new learning. Without that background, students do not just struggle to evaluate AI. They struggle to acquire knowledge, because new ideas have nowhere to land.

So we need to put the right friction back into learning. That means choosing fewer topics and going deeper, trading coverage for mastery. It also means building sticky knowledge through low-stakes, frequent routines such as retrieval practice, cumulative review, cold reads, short writes, and quick quizzes. These routines are small on purpose. Their job is to create the productive friction of recall, which makes learning durable.

AI can help here, but only if we treat it as a coach, not a crutch. An individualized AI bot can create structured friction by giving the next right question, not the next right answer. It can provide targeted retrieval prompts, spaced review, misconception checks, and feedback that sharpens thinking: “What concept are you missing?” “Where does this belong on the timeline?” “What evidence supports that claim?” Used well, it strengthens memory and builds judgment at the same time, so students are not dependent on AI output. They are equipped to challenge it. AI can assist, but students still need to carry the cognitive load.

2) Translation vs. transcreation: when the essence gets lost

AI translation removes friction fast. It reduces barriers and helps us communicate quickly. But when the friction disappears completely, meaning can flatten. A translation can be accurate and still miss the point. Tone, cultural references, urgency, and respect do not always transfer cleanly.

I saw this firsthand when a colleague asked for help translating a flyer for what was supposed to be a joyful community event. I offered to use an AI-powered tool to speed it up. It did. The translation captured the meaning, but the final draft sounded robotic. My colleague said it best: “This doesn’t feel joyous at all.” The words were technically right, but the essence was gone.

That is where we need to restore the right kind of friction. Transcreation is slower by design. It forces us to ask, “Does this sound like us?” “Would a caregiver feel respected?” “Is this the right tone for this community, in this moment?” In high-stakes communication, the right approach is humans last. Use AI to assist with a first draft, terminology, or quick options, but a human must make the final call on tone, voice, and cultural fit. If we rely only on AI translation, we risk flattening meaning and losing trust. AI can help. Humans should always own the message.

3) Sentence diagramming: making language visible again

Writing tools can make sentences appear instantly polished. That convenience can erase the friction that helps language learners understand how English actually works. Sentence diagramming is easy to dismiss as old-school grammar, and the critique is fair. Grammar drills detached from real reading and writing rarely improve writing. But diagramming was never powerful because it was traditional. It was powerful because it slowed language down and made structure visible.

I was a visual learner, and I still remember the labor of diagramming sentences. It felt like it took forever. But it worked for me. When I came to the U.S., I was in sixth grade when a teacher first asked me to diagram sentences. It helped me see English structure in a way I could not before. A sentence that felt like a blur became something I could map.

With AI, diagramming does not need to be a whole-class exercise or a worksheet. It can be an optional scaffold inside authentic writing. Visualize the structure, notice what is happening, and revise with intention. That is the friction we want. It is not punishment. It is a pause that turns language into something students can see, talk about, and improve.

4) AI-written work: bring back oral defense and process

AI can remove the friction of producing a product. That is part of the appeal. But when the product becomes frictionless, learning becomes invisible. We all know students are using AI to produce work. The issues are real: plagiarism, overreliance, and polished products that mask shallow understanding. If the response is only detection and punishment, we end up in an arms race.

A better response is to put the right friction back into assessment. Oral defense, conferences, drafts with reflection, planning artifacts, and “walk me through how you decided this” slow students down in a productive way. They do not just discourage outsourcing. They restore learning. They shift attention from the product to the process and reassert a basic principle: authorship includes accountability for thinking.

Bringing “old” back responsibly

• Bring back the purpose, not the ritual. Keep what the old practice did, drop the busywork.
• Use it as a scaffold, not a requirement. Small, targeted doses for the students and moments that need it.
• Keep it grounded in real tasks. Use it during actual reading, writing, discussion, and problem-solving, not as a stand-alone drill.

• Let AI handle the setup, not the thinking. Use AI for drafts, visuals, and prompts, while students do the explaining, revising, and defending.

The point is not nostalgia. It’s design.

I’m not arguing for a full return to the past. I’m arguing for selective recovery. Some practices were abandoned for good reasons. But others were abandoned because they were slow, inconvenient, or hard to scale, and AI tempts us to eliminate even more friction.

The challenge now is design: keep the friction that makes learning meaningful, while using AI to reduce barriers that don’t serve students. If AI is the engine, we still need a steering wheel. Sometimes the steering wheel is a slow practice: making language visible, building sticky knowledge, communicating with cultural care, and defending thinking out loud.

Borrowing from the past doesn’t make us less innovative. It might be how we stay human while we innovate.

References

WestEd. (2025). Friction by design: A framework for centering learning in the age of AI. https://digitalfluency.wested.org/resources/friction-by-design-framework/

Andy Szeto is an education leadership professor and school leader, and he serves as President of A3 of CSA. He writes and presents on instructional leadership, AI in education, and professional development, and teaches graduate courses in leadership and instructional improvement. His book, Leading Before the Title, will be released in December 2025.