Last year, I was presenting to a group of some of the most dedicated educators I have ever known. All the Teachers of the Year were brought in from around the state for a two-day conference to enhance their skills. I was invited to provide a brief overview of Artificial Intelligence (AI) to the group. Across the sessions, I noticed something unexpected ripple through the group. Shame.
There was a quiet, pervasive sense that using AI was somehow “cheating.” That if they weren’t suffering through the manual labor of lesson planning, they weren’t doing their jobs. That to be a “good” teacher meant doing it the hard way.
This resonates perfectly with a message I received from a colleague this past week. She asked:
Do you have any ideas on how to effectively utilize AI for teachers and curriculum development? I do not want them to lose cognitive stretch or the intellectual side of planning, but rather find a Goldilocks solution that cuts down on their time while preserving their creativity and expertise.
These two moments, the shame felt by the Teachers of the Year and the search for a “Goldilocks” zone, capture the central tension of our moment.
On one side, we have burnout. Teachers are drowning in paperwork and compliance. On the other side, we have atrophy (and the shame that comes with it). The fear that if we let AI do the heavy lifting, we will lose our “cognitive stretch,” that essential muscle memory of how to design a meaningful learning experience.
We need a middle ground. We need a way to utilize these tools to reclaim our time without compromising our expertise or our conscience.
The secret to the “Goldilocks Zone” is simple: Outsource the logistics, but keep the logic.
The Trap: The “Substitute Teacher” Model
The reason many teachers feel guilty about using AI, or feel like they are “cheating,” is that they often use it to replace the professional judgment required for effective teaching.
If you prompt ChatGPT with: ” Write me a 5-day lesson plan on The Great Gatsby for 10th graders,” you haven’t just left the Goldilocks zone, you’ve abdicated your role as the pedagogical expert.
In our discussions about AI supporting educators, we often get bogged down in concerns that the generated materials (lesson or unit plans, syllabi, course materials) might contain errors or hallucinations. The risk is deeper:
- It ignores Educational Psychology: The AI doesn’t know if your students are concrete or abstract thinkers. It doesn’t know their zone of proximal development. It gives you a generic average of “10th grade,” which rarely exists in reality.
- It ignores Classroom Culture: A lesson plan is not a script; it’s a reflection of the relationships in the room. Does the plan match your teaching philosophy? Does it respect the norms and trust you’ve built with your students? The AI has no concept of “the way we do things here.”
- It confuses Planning with Teaching: There is a massive difference between having a lesson plan and teaching a lesson plan. AI can generate a document that looks perfect, but it cannot help you operationalize it. Managing the pacing, the questions, the confusion, and the “aha” moments.
- It misses the Long Game: A lesson plan is just one data point. AI struggles to grasp the ebb and flow of a semester or marking period. Human educators spend years learning how to pay attention to summative assessment, which leads to formative evaluation, and how to build student growth over time.
When we use AI to generate the entirety of the plan, we become delivery mechanisms for a machine’s best guess at pedagogy. We stop being teachers and start being substitute teachers in our own classrooms.
The Solution: The “Sparring Partner” Model
To preserve “cognitive stretch,” we shouldn’t ask AI to do the work. We should ask AI to review, format, or extend the work.
Here is what that looks like across some of the regular tools we see in K-12 (ChatGPT, Gemini, NotebookLM, MagicSchool).
1. Use AI for Alignment & “Drift” Checks (The Compliance Angle)
Every educator, regardless of where they teach, answers to someone. A state board, an accreditation body, a department chair, or an institution’s strategic plan. Mapping your creative lessons to these rigid standards is high-labor, low-creativity work. This is perfect for AI.
The Strategy: The Alignment Check. Don’t ask AI to write the lesson. Write the lesson yourself. The creative, human part. Then, use the AI to ensure you are hitting the required marks.
- The Input: Paste your draft syllabus or unit plan. Then, paste the specific learning objectives (from your department, state standards, or accreditation body).
- The Prompt: ” I have pasted my draft unit below, along with the required learning objectives. Create a table showing which activities align with which objectives. Point out any objectives I have failed to address.”
- The Result: You kept the creative control. The AI handled the bureaucratic cross-referencing.
The Strategy: The “Drift” Check. It is easy to get lost in the weeds of a semester. We often suffer from “Syllabus Drift,” where our mid-semester content wanders away from our initial goals.
- The Input: Paste your entire syllabus or a long block of planning.
- The Prompt: ” Review this course plan from start to finish. Does the narrative arc make sense? Do I drift into topics that are irrelevant to the main learning goals? Identify any weeks that seem disconnected from the core purpose.”
- The Result: You get a coherence check. The AI acts as an editor, ensuring your course tells the story you intended it to tell.
2. Use Different Models for Different Goals (The “Silver Bullet” Fallacy)
One of the biggest fears educators have is that AI generates false information (hallucinates). The solution to this isn’t just “better prompting.” It is using a different architecture entirely.
The “Silver Bullet” Problem. In EdTech, there is always a desire to find the one “silver bullet” app that does everything. Recently, we have often relied on ChatGPT (or your AI tool of choice) to handle various tasks and hope for the best. But if we treat every AI interaction like a nail, we’re just hammering away with the wrong tool.
Not All Models Are Created Equal. Most of the tools we have used to date are standard Large Language Models (LLMs). Fundamentally, these are “text in, text out” engines. Usually optimized for standard academic English. They predict the next word, not the truth.
But the landscape is shifting. We now have distinct categories of models, and we need to know when to use which:
- Standard LLMs: Great for fluency, brainstorming, and drafting.
- Search-Enabled Models: Can browse the live internet for current events.
- Reasoning Models: “Think out loud” and show their logic before answering (great for math or coding).
- RAG (Retrieval-Augmented Generation): Can “read” your specific files and cite sources from your own library.
The goal isn’t to find one tool that does it all. The goal is to develop a toolkit that enables you to select the appropriate architecture for a specific purpose. (I have written about these distinctions extensively in other posts.)
For the development of teaching and learning materials, I prefer to move from standard chatbots to Retrieval-Augmented Generation (RAG) models.
What is RAG? I have written extensively about this on the blog, but simply put, RAG is like building an AI version of Google’s Custom Search Engine for your classroom. Instead of letting the model pull from the entire open internet (where it might find bad data), you force it to “retrieve” answers only from the documents you provide.
The Tool: Google ‘s NotebookLM. While you can build complex RAG systems (and I discuss those elsewhere), the easiest entry point for teachers right now is Google’s NotebookLM. You upload your specific “source of truth.” Your syllabus, readings, lecture notes, and AI answers are based solely on that context.
The Strategy: The “Tone & Tier” Check. Upload your specific planning materials, assessments, and class resources.
- The Input: Your original draft of a project description or a complex reading passage.
- The Prompt (Tone): ” Review this assignment sheet. Highlight any jargon or complex sentence structures that might be confusing for [Grade Level] students. Suggest more student-friendly alternatives.”
- The Prompt (Differentiation): ” Based on this text, create three different formative assessments: one for students who need more support, one for the standard level, and one that offers an extension challenge.”
- The Result: The AI isn’t pulling generic advice from the web; it is acting as a specialized research assistant that has read your specific files and follows your rules.
3. Use Generative AI (ChatGPT/Gemini) for “Friction Removal”
Sometimes the “cognitive stretch” snaps because we just get stuck on a blank page. At other times, the friction stems from the sheer number of hours it takes to reformat content for every learner’s needs.
AI is excellent for overcoming inertia. Both for you, as the designer, and for your students, as learners.
Strategy A: Removing Friction for You (The “Blank Page” Problem) Use AI to generate options , not decisions. When you are staring at a blinking cursor, you need momentum.
- The Prompt: ” I am teaching a lesson on quadratic equations. I’m looking for a real-world hook that will interest teenagers who enjoy gaming. Give me 10 distinct analogies or examples. I will pick the best one.”
- The Result: You remain the curator. You use your expertise to judge which analogy is accurate and appropriate. The AI just saved you 45 minutes of Googling.
Strategy B: Removing Friction for Them (Access & Differentiation) We often create materials that make perfect sense to us but create invisible barriers for students. Use AI to lower the floor without lowering the ceiling.
- The Goal: Make your language approachable and your assessments flexible.
- The Prompt (Approachability): ” I’ve pasted my assignment directions below. Highlight any jargon or complex sentence structures that might confuse a [Grade Level] student. Rewrite those sections to be more inviting and clear.”
- The Prompt (Differentiation): ” I need to assess my students on [Specific Learning Objective]. Suggest three different ways they could demonstrate mastery: one written, one verbal/audio, and one visual. Create a simple rubric for each that aligns with the same objective.”
- The Result: You are ensuring equity by offering multiple on-ramps to the learning. You aren’t “dumbing it down.” You are removing the linguistic friction so students can focus on the actual concept.
The Rule of Thumb: Logistics vs. Logic
If you are worried about losing your “humanity” or your “brain” in the process, apply this test before you prompt:
- Is this task “Logistics”? (Formatting, aligning, summarizing, brainstorming lists, checking for clarity).
- Verdict: Use AI. Save your energy.
- Is this task “Logic”? (Determining the learning objective, assessing student emotional needs, choosing the cultural context, and building the relationship).
- Verdict: Do it yourself. This is the teaching.
Closing Thought: Designing Loops Worth Living In
In the original query, my colleague mentioned the goal of “preserving their brains and humanity.”
This is the core of the issue. We often talk about keeping the “Human in the Loop” as if it’s just a safety measure. A way to catch the AI making a mistake. In recent posts, I’ve also viewed this as taking time to pause human-AI interactions and ask where the human is in this work.
But being the human in the loop is more than just supervision. It is a cognitive commitment. It means we refuse to outsource judgment, empathy, and the ultimate responsibility for what happens in our classrooms.
We don’t preserve our humanity by spending Sunday afternoon formatting rubrics or manually cross-referencing standards codes. We preserve our humanity by spending that time resting or by spending that energy connecting with students.
If AI can handle the compliance, then you can focus on the connection, and you’re not just saving time. You are designing a loop worth living in. One where the machine serves the logistics, so the human can focus on living.