Tool training stops at Substitution

Look at what districts are buying for AI professional learning this year. Prompting workshops. Tool certifications. AI literacy courses. Prompt libraries. A lot of it is good, and teachers leave with something useful: a lesson plan drafted in ten minutes, a rubric reworded for three reading levels.

Now ask what changed about the work students do.

The essay students turn in proves less than it did two years ago, and none of that training touched it.

Where the training lands

Ruben Puentedura’s SAMR model sorts technology use into four levels:

• Substitution: the tool does the same task. Nothing about the task changes.
• Augmentation: the same task, with a functional improvement.
• Modification: the task itself is significantly redesigned.
• Redefinition: the tool makes possible a task that wasn’t possible before.

Put most AI training on that scale and it clusters at the bottom. It teaches teachers to produce the materials they already produce, faster. Some of it never reaches the scale at all. SAMR was built to describe learning tasks, and this training is aimed at the teacher’s workload.

I’ve run AI workshops in many districts since 2022, and most of them started with the tools, because putting the tool in teachers’ hands is the fastest way to show what it can do. My sessions stopped there too. Teachers left with faster lesson plans and rubrics, which is exactly what the session asked them to make. Nothing in it asked what the tool meant for the work their students hand in. In one workshop, an 8th grade ELA teacher asked me how to tell whether an essay was AI-written. It was the right instinct for the task she’d been given, and nothing in the training had given her a different task.

That’s a reasonable place to start. Time is the thing teachers are shortest on, and training that gives some back earns goodwill. But the problem AI created in classrooms is that the student’s work stopped being evidence, and that doesn’t get fixed at the bottom of the scale.

The essay stopped proving anything because the task never changed, not because the teacher lacked a better prompt.

The gap is two rungs up

When a tool can produce the artifact, the artifact stops proving the student did the thinking. The fix is a task where the thinking has to show up: an oral defense, a draft history, a decision the student has to explain, a problem built so the tool gets it wrong. That is Modification, and sometimes Redefinition.

Those are design skills. They ask a teacher to decide what the task is actually evidence of, and then to rebuild it so it still is. Very little of the AI training I’ve seen this year trains for that, because it can’t be demonstrated in a 90-minute session with a login.

Here’s what that looks like on one task. Before: a five-paragraph argument on a teacher-chosen prompt. A weak redesign adds a reflection paragraph at the end, which the tool can write too. A stronger one keeps the argument and adds a three-minute conversation where the student defends the counterclaim they chose to leave out.

Why tool-first training keeps resetting

Tool-first training has a built-in reset. Tools turn over every few months. A district that trains around a tool retrains every time the interface, the model, or the license changes. Each cycle starts teachers back at the beginning: What is this? How does it affect me? In Hall and Hord’s Concerns-Based Adoption Model (CBAM), those are the early stages of concern, informational and personal. The later stages ask different questions: what is this doing to my students, and how could it work better? That’s where practice changes.

Tool training adds a stop in between, the stage CBAM calls Management: how do I fit this into my day? A workshop on faster lesson plans answers that question well, and that’s where it leaves people.

Tool training keeps pulling people back to the early stages. The tool is new again, so the concerns are new again. The change never settles long enough for teachers to get to the question that matters.

A skill like “redesign a task so the thinking shows” doesn’t expire when the tool updates. It carries forward.

What training for the upper rungs looks like

It looks less like a workshop and more like the work itself:

1. Start from a real assessment. Bring one common task the department already gives.
2. Place it on SAMR, honestly. Where does the task sit today? Could a student complete it with the tool and learn nothing?
3. Score it together. Three teachers, five pieces of student work, blind. Where they disagree is where the task is unclear about what it’s evidence of.
4. Redesign one step up. Not to Redefinition on day one. Modify one element so the thinking has to show.
5. Come back in six weeks. Score the new version the same way. Then look at how teachers are using the redesign, with CBAM’s Levels of Use: still following the steps (Mechanical), doing it as a matter of course (Routine), or adapting it for their students (Refinement).

That’s a cycle. It changes behavior because it asks teachers to do the new thing with their own students’ work, and then shows them whether it worked.

A cycle like this needs someone to run it: to protect the time, hold the six-week return, and keep the conversation on student work when it drifts back to the tool.

A caveat about SAMR

SAMR is a lens, not a ladder. It was never validated as a research instrument, and “higher” isn’t always better. Plenty of good teaching happens at Substitution. The point isn’t to push every task to Redefinition. It’s to notice that the training stops below where the problem lives.

Where to start

Ask your PD lead one question: of the AI training we’ve delivered this year, how much of it changed a task students complete? If the answer is “not much,” you don’t need to cancel anything. You need one department, one assessment, and one cycle aimed one rung higher. If you need help getting started, call me and I’d be glad to work with you on a plan.