Let Them Learn Where Mistakes Are Cheap

A finished practice app can hide an unfinished lesson.

If you’re leading a team learning to build with AI, you can look at the screen and see progress. There are buttons. The information appears. Something exists that didn’t exist before.

Then you consider the next assignment. What did this person learn to decide that makes the next responsibility a reasonable choice?

That answer may be harder to find than the app.

The internal-app roadmap behind this piece had future ideas waiting in a queue and one next project selected for activation. That gave the work an order. It was still a plan. It did not establish what had been built, what the learner could correct, or what responsibility they could carry next.

Here’s the thing. You can plan a sequence of projects without planning a single decision for the person doing them to own.

Give them a real decision with a recoverable consequence.

Find the choice inside the assignment

Imagine a colleague making a practice version of an equipment borrowing list with AI. All the items are invented. So are the borrowers. Nobody is using it to find a real drill or reserve something for tomorrow’s work.

The useful result is simple: another person should be able to tell whether an item is available.

You could hand over a detailed set of instructions. Put the name here. Make this button green. Add this label. Follow these steps until the screen looks like your sketch.

They might follow every instruction and produce exactly what you asked for. You would know they could work through that assignment. You might still have learned very little about a choice they made themselves.

Instead, leave them something real to decide. How should the list make availability clear when an item is borrowed and then returned?

They can use AI to explore approaches. They can ask you questions. They still need to choose an approach that makes sense for the person reading the list.

The assignment has a useful job and a decision inside it. Both matter. A blank invitation to build anything can leave the learner guessing what you value. Instructions that decide everything can leave them with nothing to judge.

Cheap needs a definition

The Prep List gives us two useful questions for choosing this kind of assignment: Definability and Reversibility.

Definability asks whether you can describe a useful result. For our practice list, someone can tell whether an item is available. A lovely screen showing contradictory answers has missed that result.

Reversibility asks how hard a wrong result would be to repair. In this exercise, you can correct or replace the practice copy before anyone relies on it. No real borrower is sent looking for equipment that isn’t there.

Those questions help you choose the assignment. They don’t give you a score for the person.

And internal doesn’t automatically mean cheap. A wrong answer in an internal list can still waste a colleague’s morning. A practice version with invented information gives you a place to examine the choice without handing that consequence to the team.

Yeah. You still spend time reviewing it. The learner still spends time repairing it. Cheap means the consequences are manageable, not that teaching costs nothing.

Agree on the useful result, where the practice work stays, and when you’ll look at it together. Name someone the learner can ask for help. If you can’t keep a mistake manageable, make the assignment smaller before starting.

Stop before you supply the answer

Stay with our hypothetical list. Suppose an item now appears as both borrowed and available.

That’s a useful moment to slow down. The screen gives another person two incompatible answers. Something about the chosen approach allows that contradiction.

Your first instinct may be to fix it. You know what you would change. You can probably describe the correction faster than the learner can explain how they got here.

Hold that answer for a moment.

Ask them to show what they expected to happen when the item was borrowed. Then look at what actually changed. Where did the two answers come from? Which part of their approach allowed both to remain visible?

You’re listening for a connection to the work in front of you. A fluent explanation from AI might help them investigate. It doesn’t answer the question by itself. Can they point to the choice and show why it produced this result?

Don’t assume the cause before you’ve looked either. Perhaps they treated the two labels as separate things to update. Perhaps the intended change never happened. Those would need different corrections.

The learner may find the problem. They may need a hint about where to look. They may need you to explain something they haven’t understood yet.

All three are useful things to discover before the next assignment.

Help without hiding the lesson

Here’s the thing. Withholding help isn’t the same as giving someone a decision to own.

You don’t need them stuck for an afternoon to prove they’re trying. Agreeing on a review point also means agreeing that questions are allowed before then.

If they can’t find the contradiction, you might draw their attention to the two labels. Then let them work through what each one means and what should happen when the item is borrowed.

If you need to explain the whole relationship, explain it. That tells you what still needs teaching. Quietly fixing the list yourself and calling the assignment complete would hide that information.

This isn’t a contest between the learner and AI, either. They can ask AI for possible causes or a proposed change. What matters is whether they can judge the suggestion against this list and show what the change does.

A proposed repair is a useful next step. Actually. The repair still has to happen.

Have them make the change in the practice copy. Then show what the list says when the item is borrowed and when it is returned. Has the contradiction gone? Can another person now tell whether it is available?

Changing one label by hand might make this screen look right while leaving the next borrowing action to create the same problem. Ask them to show how their correction handles that action. That keeps the review connected to the choice you wanted them to examine. If they can only explain the screen after you have repaired it, you still have a teaching need to work on together.

Keep track of the help that made that possible. A correction made after one hint and a correction you largely supplied leave you with different teaching decisions.

Does that make sense? You’re trying to see where the learner can carry the thinking and where they still need support.

The next assignment follows what you saw

Suppose they can explain the choice, find the contradiction, and demonstrate a correction. You have something specific to build on. The next bounded assignment might give them another choice of similar consequence.

Suppose a hint gets them moving. You might stay at this level and give them another chance to recognize the same kind of problem.

Suppose the relationship between borrowing and availability is still unclear. Make the assignment smaller. Work through that relationship together before adding more features.

Those are possible outcomes of our invented exercise. None is a verdict on the person’s worth. And none follows automatically from how finished the app looks.

One correction also doesn’t tell you how they’ll handle every problem. It doesn’t establish that they’re ready for customer work or that supervision can disappear. It gives you a reason for the next learning choice, with its limits attached.

If nothing naturally goes wrong, don’t manufacture a surprise failure. Agree openly on a practice discrepancy and have them demonstrate how they would inspect and repair it in the practice copy. The purpose is to see their reasoning, not catch them out.

Your Monday Move

Choose one learner and one small internal practice assignment. Write down the useful result and the decision they will own. Use approved, non-sensitive practice information and keep it separate from work people depend on.

Agree on what a mistake could affect, how the practice work could be restored, where help is available, and when you’ll review it together. If the consequence is too large, shrink the assignment.

At review, ask them to explain one choice and demonstrate a correction in the practice copy, using an agreed discrepancy if needed. Notice what they can show, the help they need, and anything still unresolved.

Then choose what comes next for a stated reason: another attempt at this level, a smaller assignment, or another bounded challenge.

The roadmap can wait for that answer.