A session with Karen Levin, founder of Math for Humans | Presented by EdTechTeacher
Every math teacher knows this reality: one class, one period, students at wildly different places. You have confident problem solvers, multilingual learners, students carrying math anxiety, and students who need an extension before you have even introduced the concept. Differentiating for all of them, on the same day, every day, is one of the most demanding parts of the job.
AI tools are being marketed as a solution to this and nearly every other challenge in education. But the more useful question is not whether AI can help. It is whether it can help without lowering the bar for student thinking, and whether you know when to put it down.
In a recent EdTechTeacher webinar, Karen Levin, founder of Math for Humans and a nationally recognized math educator, walked through exactly that. This post captures the key ideas from that session, with practical moves you can bring to your own classroom or coaching work.
VIEW THE FULL WEBINAR HERE or EMBEDDED BELOW
Karen opened the session with a math task: a fraction area problem about a flag replica. Participants solved it, shared their strategies, and examined student work samples showing a range of approaches: correct, partially correct, and misconception-laden. Then Karen did something instructive. She handed the same problem to a chatbot.
The AI got it right immediately. Every time. That is not a crisis, but it is useful information. Karen frames this as an assignment audit: testing what an AI tool can do with a task before you assign it.
A few questions worth asking about any task:
– Could AI easily complete this task?
– Would most students produce nearly identical work?
– Does it lack meaningful voice, choice, or creativity?
– Could it be framed in a more real-world or open-ended format?
The audit is not about eliminating AI. It is about understanding where student thinking is still protected, and where a redesign might be worth the effort.
When Karen asked Gemini to “differentiate this task” with no additional context, she got what most educators get: a mixed bag. Some ideas were useful. Some were generic. Some actually lowered the cognitive demand of the original task.
The tool was not being lazy. It just did not know what good differentiation looks like in a math classroom. So Karen added a lens.
First, she prompted with the CAST Universal Design for Learning (UDL) guidelines: research-backed principles for designing multiple means of engagement, representation, action, and expression. The output improved. Then she added the NCTM Five Practices framework (anticipate, monitor, select, sequence, connect), and the results became math-specific and substantively more useful.
“It’s one class, many entry points, and a shared mathematical conversation.” — Marian Small, math education researcher
The core takeaway:
“We’re using AI to help us think more carefully, not to think for us.” — Karen Levin
Karen also pushed the prompt further, asking the AI to identify what might not work, what teacher moves would be needed to make a suggestion realistic, and how to connect the task to specific student interests. That iterative pressure, not accepting the first polished answer, is where the real work happens.
One important caveat: these tools are sycophantic by design. They want to please you. If your plan has gaps, the AI will often validate it anyway. Stay grounded in your own professional judgment. The goal is to have the tool challenge your thinking, not confirm it.
Entering long prompts every time is not sustainable. Karen’s solution is to build a “gem” (Gemini), a project (ChatGPT), or a folder — the naming varies by platform, but the concept is the same. You configure it once with your instructions, a knowledge base, and an output format. From then on, every interaction pulls from that context.
For differentiation, Karen’s gem includes the core strategies from a math-specific research framework, step-by-step instructions for how the AI should approach a new task, and clear rules about maintaining cognitive demand. She also attached the actual curriculum materials she references, so the AI pulls from a known source rather than the open internet.
The Microsoft Work Trends Index 2025 describes three phases of AI use: AI as assistant (you prompt it, it responds), AI as digital colleague (it works under your direction on repeatable tasks), and AI as autonomous agent. For most educators, the goal is phase two. If you have something you do repeatedly, a configured project is worth building once rather than re-entering the prompt every time.
Karen has personally reviewed more than 25 math AI tools. Her consistent finding: there is no silver bullet. But there is a useful distinction between two types of student-facing tools.
“Step givers” walk students through solutions step by step. They are effective for practice and procedural fluency work, but they risk having students mimic rather than think. “Sense makers” ask students to explain, reflect, and justify their reasoning. They are better for reasoning development, but students who do not know where to start may simply shut down.
The fix for step givers: build classroom routines that explicitly position students as thinkers, so the AI is a resource and not the authority. The fix for sense makers: give students a clear routine for what to do when they get stuck.
“How does a tool position our students as mathematical thinkers?” — Karen Levin
Karen shared three examples of student chatbot use that maintain that positioning:
1. Remove the numbers from a task and have students figure out what information they need before solving.
2. Have students solve a problem, then ask the AI to solve it, and compare — this encourages skepticism and error-detection.
3. Prompt the AI to roleplay as a historical mathematician and have a conversation about a concept, then share that transcript with the teacher.
In each case, the student is still doing the cognitive work. AI is a thinking partner, not a shortcut.
Karen also flagged something worth taking seriously: data shows students are already using general chatbots regardless of what schools tell them. A culture of transparency, clear expectations for when and how AI is appropriate, and honest conversation about the risks, is more effective than blanket restrictions. AI detectors, she noted, are unreliable and often produce false positives.
Before adopting any AI tool or approach, Karen’s flowchart is a useful filter:
1. Do you have a clear problem to solve? If not, stop. AI for the sake of AI wastes time.
2. Could AI meaningfully help with that specific problem? Sometimes the answer is no.
3. Is this for teacher use or student use? Get clear before you proceed.
4. Does the tool deepen student reasoning or replace it? If it replaces reasoning, it is not the right moment.
5. Is the improvement worth the cost? Factor in time, training, and any financial investment.
This is not a framework for avoiding AI. It is a framework for using it purposefully.
AI is most useful in a math classroom when the educator using it knows their content well enough to judge what the tool gives back. A weak content lens means a polished but rigor-lowering output can slip through. A strong content lens means you can spot the gaps, take the useful ideas, and move forward.
Karen’s framing is worth keeping close: AI should act as a critical partner that pushes your thinking, not a machine that produces finished work. The thinking, the judgment, and the relationships that drive real learning in math classrooms still belong to teachers and students.
Watch the full session on the EdTechTeacher YouTube channel: https://youtu.be/sEEkq6KCjL8
Explore upcoming EdTechTeacher events: https://edtechteacher.org/events/
Sign up for the EdTechTeacher newsletter: Newsletter
About the Presenter
Karen Levin is the founder of Math for Humans and a nationally recognized educator with more than 15 years of experience at the intersection of math instruction and educational technology. She has coached educators and administrators at more than 20 schools. Her work centers on a simple but powerful principle: AI should support student reasoning, not replace it.
About EdTechTeacher
EdTechTeacher helps K-12 educators and instructional leaders use technology to transform teaching and learning. We offer workshops, institutes, and online courses for teachers and schools.