Show your work, even when it hurts

The sensible thing to say about artificial intelligence is that we should learn how to use it. That is also the boring thing to say, although boring things often have the annoying habit of being true.

The harder thing to say is that we should also keep doing some work the slow way.

I was thinking about this after watching Flux Academy’s video, “The real problem for Web Designers using AI”. The video is aimed at web designers, but the point is bigger than design. The danger is not simply that a client, a manager, or a student can type a prompt and get something that looks finished. The danger is that enough “finished-looking” work can make us forget what skill feels like while it is still being built.

Students are not wrong to use tools. I use tools. I teach tools. A calculator is a tool. A compiler is a tool. Google is a tool. The textbook is a tool, although it is not always loved as one. The problem comes when the tool does the thinking before the student has learned what thinking in that field is supposed to look like.

This is why “show your work” is not just a fussy teacher phrase from the chalk-dust era. It is the thing that reveals whether a person knows where they are, where they are going, and why the next step belongs there. In math, that might mean writing the steps. In programming, it might mean sketching the logic before asking the machine to produce syntax. In writing, it might mean leaving the outline, the bad first paragraph, and the source notes visible long enough to learn from them. In design, it might mean wireframes and rejected versions before the polished mockup shows up to take a bow.

The old-fashioned part matters because friction matters. Not all friction, of course. Some friction is just bad software, bad instructions, or somebody making a class harder because they survived it that way and now feel history should repeat itself. But some friction is useful. Robert and Elizabeth Bjork’s work on “desirable difficulties” is a good reminder that the thing that feels slower in the moment can build more durable learning. Retrieval practice, spacing, interleaving, explaining your reasoning: these are not always efficient in the short term, but education is not supposed to be a race to the nearest answer-shaped object.

There is also a reason worked examples and self-explanations keep showing up in learning research. Chi and colleagues’ classic work on self-explanations points to something teachers see all the time: students learn more when they have to explain why a step works, not just copy the step. Atkinson, Renkl and Merrill’s work on fading worked-out steps gets at the same neighborhood. First we show the road. Then we remove a few signs. Eventually the student has to drive.

That is the part AI complicates. It can produce the road, the signs, the car, the soundtrack, and a confident explanation of why the trip was necessary. Sometimes it is right. Sometimes it is almost right in the most dangerous way. Either way, the student who never had to struggle with the map may not know the difference.

So yes, students need to show their work. They need notebooks, sketches, drafts, commit messages, pseudocode, margin notes, attempts that fail for specific reasons, and revisions that show what changed. They need to practice doing things the old-fashioned way long enough to know what the machine is doing when it helps and what it is hiding when it cheats them out of the learning.

But instructors have to endure the pain too.

That may be the more uncomfortable part. It is one thing to tell students not to outsource the struggle. It is another thing to design a class where the struggle is visible, useful, and graded without turning the instructor into a paperwork-processing machine. If students have to show process, instructors have to look at process. If students have to revise, instructors have to make room for revision. If students have to explain where an AI answer helped or failed, instructors have to read those explanations and resist the lazy comfort of grading only the shiny final product.

That means our assignments need to change. Not because every assignment must become an anti-cheating fortress, but because the finished artifact is now too cheap a signal. A polished answer may tell us less than a messy trail. The trail is where judgment appears.

This is not new. The idea of cognitive apprenticeship has been around for decades: experts should model the work, make their thinking visible, coach students through practice, and gradually remove support. That sounds old-fashioned because it is old-fashioned. It also sounds like teaching.

The twist is that instructors need to show their work as well. We should let students see how we approach a problem before it becomes a rubric item. We should solve something live and make a few mistakes where they can see them. We should talk through why an AI-generated answer is plausible but weak, or efficient but shallow, or technically correct but useless for the actual audience. We should be honest that this takes time and that time is the thing everyone is trying to save.

The Flux Academy video argues, in effect, that creative people still matter when they protect the part of the work that requires taste, judgment, and expertise. I would add that students do not get those qualities by receiving only completed answers. They get them by wrestling with process until the process starts to become their own.

And instructors do not get to skip that part either. We have to build the kind of classes where the work is visible enough to teach from, painful enough to matter, and humane enough that students understand the pain is not the point.

The point is learning.