When AI Breaks Your Assignments

It's time to get better at using AI

The Second Draft: #0087

I write weekly articles for educators who are ready to get unstuck from outdated curriculum, resistant institutions, and a career that was built for a world that no longer exists.

---

Where We’ve Been

Over the last few posts, we’ve been combating probably the most common misconceptions about education:

that student brains are like filing cabinets just waiting to be filled with facts (that, btw, we uniquely have access to 😆).

To combat this terrible lie,
we introduced a formula about how learning actually works.

(I’m actually updating this slightly here for simplicity, JICe you were paying close attention)

Learning = WM_{i-1} × (GC × EA × TC_i × DP)

We broke down each component:

• WM: World model

• GC: Genuine challenge

• EA: Embodied agency

• TC: Threshold concepts

• DP: Deliberate practice

Practically this is what this looks like:

You have a world model (WM) of how the world works—which is basically just an over simplification of all the potential truths, possibilities, and probabilities of any situation—that happens to work well enough that you are able to function.

A genuine challenge (GC) is anything that disrupts your WM (positively or negatively) and is important enough for you to try to resolve it.

Resolution flows through the embodied actions (EA) you take, which are basically choices over which you have enough control that you actually feel the consequences of those decisions (physically, cognitively, emotionally, etc.).

Those choices can become threshold concepts (TC), which are themselves simplifications of phenomena, best guesses about the smallest thing you can do or think about—we can think of them as either intuitions, or more formally as heuristics or rules of thumb—that allow you to move closer to stabilizing your WM.

These TCs are then put to the test through deliberate practice (DP), determining which ways of understanding the world work and which don’t (both have value within a WM).

And as you find the TCs to keep and reject, you end up with an updated WM.

Repeat, ad infinitum.

---

The Formula Workflow

Okay, let’s put the formula into action.

Here’s how:

Step 1: Decompose the Activity

This is the most important part of what we’re about to do, so don’t skip it.

Decomposing an activity means identifying what knowledge, skills, and abilities (KSAs) your assignment actually requires and makes possible.

If we’re honest, we don’t really know so we’ll need to ask AI to help us breakdown our activities.

Specifically, we want to map the activity along 4 levels:

Foundational: Basic procedures, skills, knowledge needed to complete the task

Analytics: Ability to distinguish, recognize patterns, identify relationships

Synthetic: Integrate multiple elements, build coherent wholes, make strategic choices

Epistemic: Understand the nature of knowledge in this domain

Doing this is basically like asking AI:

What KSAs do you suppose I was trying to get my students to develop and demonstrate when I assigned this [insert your assignment]?

Actually using AI to do this is alone worth reading this entire series, IMO.

You will probably be very surprised by what your simple assignment both requires and makes possible.

Let’s see what this looks like with an example we’ll use for the rest of this article.

Example: 90-Second Pitch Assignment

(Note: We picked a relatively interesting assignment to show that we can always improve. Even assignments that seem engaging can be dramatically enhanced.)

The original assignment: “Pick a side of a topic of your choice and record yourself making a 90-second pitch for it.”

AI-generated decomposition:

• Foundational

  • Record video or audio

  • Speak clearly on camera

  • Hit the time constraint (90 seconds)

  • Select a defensible position

  • Use basic argument structure (claim + support)

• Analytical

  • Identify the strongest argument for the position

  • Distinguish persuasive evidence from weak support

  • Recognize what moves an audience versus what just informs

  • Adapt the message to the implied audience

• Synthetic

  • Build a tight case from multiple supporting elements

  • Anticipate and preempt counterarguments

  • Structure the argument for impact under a severe time limit

  • Choose what to emphasize versus omit strategically

• Epistemic

  • Understand persuasion as rhetorical construction (not just stating truth)

  • Recognize that every argument makes strategic framing choices

  • Understand that taking a “side” requires recognizing that other sides exist

See how much more the assignment could be teaching than you originally thought?

You can also do this down as many levels as you like. But, one level is typically good enough.

Step 2: Pick the KSAs That Matter Most to You

Since the AI is just guessing what you might have wanted to achieve, you might need to do some tweaking on the step above, but in any case you definitely need to do some selecting on this step for the KSAs most relevant for your goals.

And to make this as developmental as possible to fit with our formula, in an ideal world you’re picking at least one KSA from each level.

The KSAs we picked for the 90-second pitch:

• Foundational: Basic argument structure (claim + support)

• Analytical: Recognize what moves an audience vs. just informs

• Synthetic: Anticipate and preempt counterarguments

• Epistemic: Persuasion as rhetorical construction (not just stating truth)

Notice: We picked one from each level such that each one requires the previous level as foundation.

---

Step 3: Create a GC That Connects Them

With our KSAs in placed, and laddered, you can have AI propose a sequence from the lowest to the highest level.

Note that each part of the task is actually a GC (a genuine challenge that disrupts their current world model).

Also note that each subsequent step does two things:

1. Introduces a new GC

2. Encourages reinterpretation of the previous WM that the prior resolution produced

In this sequence, the initial challenge is disrupting the existing WM.

And each subsequent step acts as feedback that disrupts the WM that preceded it. This is exactly what we need to happen for learning to happen—GCs that encourage EAs producing TCs developed through DP, that reshape our WMs!

The 4 parts of our redesigned pitch assignment:

GC_L1: Build Your Structure (targets Foundational)
Make your 90-second pitch. Try different structures. Pick the one you think works best.

GC_L2: Test If It Moves (targets Analytical)
Pitch to 3 people. After each one, they say one word: “Informed” or “Moved”

The break: Your structure holds together BUT you only inform, don’t move.

TC surfaces: Structure ≠ persuasion

Rebuild with stakes/urgency/emotion.

GC_L3: Get Attacked (targets Synthetic)
Present your persuasive pitch to someone who disagrees. They attack your weakest point.

The break: You moved them BUT left yourself exposed.

TC surfaces: Should have addressed that objection preemptively.

Rebuild with counterargument in opening 30 seconds.

GC_L4: Reverse the Position (targets Epistemic)
Now argue the opposite side using the same persuasive techniques.

The break: You realize your “truth” was actually a constructed argument.

TC surfaces: Persuasion is rhetorical construction, not just stating facts.

Build the counter-pitch using the same structure/moves that made your original persuasive.

---

Do you see how that builds?

Did you notice how each step develops some KSAs, but that each subsequent challenge challenges what was just built such that new KSAs need to be called on to resolve the challenge?

This is the entire point!

---

The Two Essential Pieces

That’s not the entire workflow.
There are two parts that are absolutely essential:

First: PREDICT

Before each attempt, the first things students do is basically answer:

What do you think you will need to make this work?

Using are metacognitive framework,
here are 5 questions you can use as is for any assignment:

1. Key Terms: What are [#] big ideas, skills, principles, or ways of thinking you’d need to [challenge phrase]?

2. Working Definitions: Pick the [#] from your list that you think matters most. Explain how it actually works — like you’re teaching it to someone who’s never done anything like this before.

3. Anchor Moves: If you were watching someone really skilled [challenge phrase], what specific moves would you see them making? What are they looking for?

4. Quality Signals: What would tell you your attempt is actually working — not finished, just on the right track?

5. Break Points: Imagine [challenge phrase] is going well . . . and then something goes wrong. What do you think broke?

Do you see what’s going on with these questions? Students are essentially uncovering, as best is possible, the WMs they have about this challenge. And their predictions actually become GCs that will support further development as students get to the second piece. 👇

Second: UPDATE

After each attempt, students answer these 3 questions:

1. If your future-self watched you completing that, what 3 things would they say you missed?

2. On a 1-10 scale, how strong was your work? What makes it that number instead of one lower?

3. Imagine you improved this next attempt beyond what you expected. What were you doing that made that possible?

This isn’t metacognition for its own sake.
It’s one of the most powerful ways to boost memory.

Research shows that making a prediction—even an incorrect one—creates a much stronger memory of the actual outcome than if you just observed it or read about it!

These predictions and updates also serve as an instant feedback source to students, given to them by themselves! Bonus! 🎁

---

The Complete Workflow (worked example)

Okay, here’s what this looks like from the perspective of a student doing the work.

You decompose your assignment.
You get your KSA table.
You pick the ones you want.
You create the 4-level process.

And then students:

PREDICT →
5 questions, focused on: What structure will make your pitch persuasive?

GC_L1: Build Your Structure
Make your 90-second pitch. Try different structures.

UPDATE →
3 bridge questions

PREDICT →
5 questions, focused on: What will happen when real people hear it?

GC_L2: Test If It Moves
Pitch to 3 people. They say “Informed” or “Moved”

UPDATE →
3 bridge questions

PREDICT →
5 questions, focused on: What objection will someone raise?

GC_L3: Get Attacked
Present to someone who disagrees. They attack.

UPDATE →
3 bridge questions

PREDICT →
5 questions, focused on: Can you argue the opposite side persuasively?

GC_L4: Reverse the Position
Argue the opposite side using the same techniques.

FINAL UPDATE →
3 bridge questions

---

Why This Works

Now that is pretty awesome, isn’t it?

We turned one activity, into a compounding feedback loop that was driven entirely by the experience itself and the student’s own metacognitive engagement with it!

Mission accomplished, I’d say!

Here’s why this is so powerful.
Each iteration:

• Builds on the previous level

• Breaks the student’s current world model through feedback via the next level of the challenge and their own metacognitive reasoning about it

• Forces new threshold concepts to surface

• Requires rebuilding their world models with new clarity of insight

The formula activates at each level:

Old WM × (GC breaks it × EA through doing × TC emerges × DP through rebuilding) = New WM

It’s ONE assignment that develops multiple capabilities through designed GCs.

Go us!

---

Next Time

This workflow works, as is.

You can use AI right now to decompose any assignment, select targets, and design the ladder.

But next week, I’m going to share the files and the complete bot system that does all of this automatically.

You’ll paste in any assignment.
The bot will:

• Generate the decomposition table

• Let you select target KSAs

• Design the laddered challenge with all breaks built in

• Output the step-by-step procedural guide

The entire process takes 5 minutes.

I built it, you get to use it and refine it!

See you next week.