What Happens When AI Learns to Make My Sauce?
My family has a spaghetti sauce recipe that’s been passed down for generations. I make it. My mom makes it. My uncle makes it. My nonna made it. We’re all technically following the same recipe, but none of our sauces taste exactly the same.
Part of the problem is that the recipe isn’t really a recipe. It’s a handful of this, a pinch of that, cook it until it looks right, taste it, add a little more of whatever it needs, and stop when it’s done.
What’s a handful? How much is a pinch? What does right look like? How do you know when it’s done?
Those answers come from making the sauce. They come from watching someone else make it, making it badly, tasting it, adjusting it, and eventually reaching a point where you don’t have to think about every decision. You just know.
I could also buy sauce from the store.
Honestly, in a blind taste test, the store-bought sauce might hold up. It might be just as good. It might be better. But I’d still know one came from a jar and one was made by somebody standing over a pot.
Why does that matter?
I don’t know yet.
Teaching the Handful and the Pinch
I’ve been using AI to help grade design assignments this semester. That has created its own set of logistical questions about cost, efficiency, accuracy, and whether I’m slowly training a computer to take my job. But it has also forced me to do something I hadn’t expected: explain how I judge design.
Not just the rubric. The judgment underneath the rubric.
Why does this hierarchy work? Why is that button state too quiet? Why do these elements technically share a style but still not feel like a system? Why is one problem a small refinement while another one means the whole design isn’t functioning? Why does breaking a convention work here but not there?
AI wants me to quantify a handful and a pinch.
At first, that seemed like the obvious place it would fail. I’ve been working in visual communication for a long time. Some of these decisions live in my gut. I look at something and know it isn’t working before I’ve put words around why.
But then I give the AI examples. I correct it. I explain what it overvalued and what it missed. I show it when a problem matters and when it’s being too picky with a beginner. I give it more context about the assignment, the students, and what we talked about in class.
And it gets better.
Sometimes uncomfortably better.
The more this happens, the more I wonder if design and creativity follow more of a recipe than we like to admit.
We teach hierarchy, contrast, legibility, scale, composition, pacing, semiotics, color relationships, typographic associations, and motion principles. We tell students that form follows function. We give them rules and then teach them when those rules can be broken. We critique their work, they try again, and eventually some of those decisions move from conscious effort into gut reaction.
Commercial design has even more structure around it. There’s an audience, a message, a client, a medium, a deadline, and usually somebody paying for a particular result. We may approach the problem creatively, but we’re still working inside a box.
That doesn’t make design easy. Making good sauce isn’t easy either. But it may make design more teachable, explainable, and reproducible than we want to believe.
So is creativity really as valuable an asset as we’ve treated it? Or have we been using the word creativity to describe a complicated pile of craft, judgment, taste, experience, and rules?
Maybe Nothing Is Original Anyway
I don’t really believe in originality, at least not in the romantic idea that something appears from nowhere.
Star Wars is Kurosawa in space, mixed with a dash of pulp science fiction and a handful the hero’s journey. That doesn’t make it less interesting. It means we can see parts of the recipe.
When I worked as an animator in video games, I made a point not to study other video-game animation. I watched ballet. I watched sports. I looked at movement that wasn’t part of the genre I was working in, then brought those references back into the characters I was animating.
I ask students to do something similar when they research a project. Find something exactly like what you’re making. Find something similar. Find something related. Then find something completely unrelated. That last reference is often where the interesting part comes from.
I also do an exercise where students create separate columns of related words and then quickly draw lines between the columns. They make connections before they have time to decide whether the connections are good.
Most of them aren’t.
Randomly connecting two words isn’t especially creative. The useful moment happens when someone looks at one of those collisions and thinks, Wait. There might be something there.
Maybe creativity isn’t making the connection. Maybe it’s recognizing that the connection matters and deciding to follow it.
That sounds a lot like judgment. Knowing when to stop, taste the sauce and decide how to move forward.
And that’s exactly what I’ve been trying to teach AI.
Gut Reaction Might Be Compressed Experience
When my nonna said a handful, she wasn’t invoking some unknowable force. She’d made the sauce enough times that the measurement lived in her hand.
Maybe gut reaction works the same way.
My design judgment contains years of good projects, bad projects, client notes, critiques, technical mistakes, abandoned ideas, cultural references, deadlines, and tiny decisions I don’t remember making. I don’t consciously retrieve all that information when I look at a student’s work. It’s already been compressed into the feeling that something is wrong.
Teaching AI how I grade means unpacking that feeling. This is too much. This isn’t enough. This technically follows the rule, but it doesn’t work in this context. These two problems look similar, but this one matters more. This student is a beginner, so don’t grade the work like it came from a professional design studio.
I mean, that’s how we teach, right? Students make something, get corrected, try again, and slowly develop taste. They learn the recipe, then they learn what a pinch means to them. That’s just reinforcement training, isn’t it?
AI can make connections. I can tell it to look somewhere unexpected, combine unrelated references, or generate far more possibilities than I could explore myself. I can also show it which possibilities I accepted, which ones I rejected, and why.
Does that mean AI recognizes when a connection matters? Or does it become extremely good at predicting what I'll recognize as meaningful?
I don’t know how different those things are in practice.
There are also ethical questions here about training data, labor, energy, ownership, and who profits from the system. Those matter. A thing can be creative and still come from an exploitative system. Calling it uncreative doesn’t solve the exploitation, and calling it creative doesn’t excuse it.
But those are different question to stir into another recipe.
The Jar Looks Pretty Good
The bigger problem for my students may be that the first result usually isn’t bad.
The sauce from the jar tastes fine. It might look polished. It might be better than what a student thinks they could make on their own. So why keep going?
Why generate another direction? Why ask what’s missing? Why tear apart something that already looks presentable? Why learn how the thing works if the result arrived before you understood the process?
I can delegate parts of a project because I have enough experience to recognize when the result is wrong. If AI invents visual evidence while grading, I can catch it. If it applies a convention too rigidly, I can tell it that the convention doesn’t matter here. I know what failure looks like because I’ve failed at these things myself.
A beginner doesn’t have that backlog yet.
That doesn’t mean students need to recreate every old production process forever. The process has changed. Pretending AI doesn’t exist won’t prepare them for anything. But a new process still needs somewhere for curiosity to happen.
Students need to learn not to accept the first plausible answer. They need to ask questions, try something they aren’t sure about, follow a bad idea far enough to find the good part inside it, and discover which parts of making they actually enjoy.
I don't think what's being lost to AI is creativity. It's gotta be what's lost in the experience that teaches someone how to recognize creativity.
I’ve noticed a version of that loss in my own AI-assisted grading. I still read and review every student’s work and edit much of the feedback before a grade is posted, but I don’t know the projects quite as intimately as I did when I handled every part manually. When a student approaches me, sometimes I need a moment to remember which project is theirs.
The grading may be accurate. The feedback may even be better. But something changed in my relationship with the work.
Again, I don’t know exactly what that means.
The Parts I Like
I tell students that style isn’t necessarily making everything look the same. Unless you’re one of the relatively rare people hired to repeat a signature look, professional creative work requires adapting to different clients and contexts.
Style may come from the parts of the process you repeatedly enjoy.
When I make motion graphics, I love a very particular easing curve. It eases in and out in a way that just feels good to me. Most clients would never identify it, but I use it constantly.
I’ve got a salmon-colored solid layer I like to place over projects with a particular blending mode. Turned down low, it creates a warm ambient glow that somehow works even when the rest of the project is cold.
I love building procedural systems that anticipate client changes. If I know someone may want to change an early decision, I try to construct the project so that changing one thing flows through everything else without forcing me to rebuild it. That feels like solving a puzzle.
Those are tiny things. They aren’t necessarily visible in the final result. They’re just parts of making that make me happy.
AI doesn’t reproduce all of them yet because I haven’t asked it to. I haven’t explained the curve, the glow, or the exact way I like to organize a project.
But I’m sure I could.
Then what?
Do I keep doing those things because I enjoy them? Do I let AI do them because it can? Does my preference move somewhere else? Does style become the way I direct, select, reject, combine, and revise instead of the way I move the curve myself?
The point is using AI creates a new process; it isn’t identical to the old process.
I can follow the family recipe and make a sauce that tastes different from everyone else’s. I can buy a jar that probably tastes pretty similar. I can modify the jar. I can teach someone the recipe. I can finally measure the handful and write down the pinch.
Eventually, AI may be able to reproduce my version perfectly.
And I may still want to make it myself.
Or maybe I’ll discover that the part I loved wasn’t standing over the pot. Maybe it was adjusting the recipe, remembering where it came from, making it to someone, enjoying people taste it, or figuring out what the rest of the meal needed.
I don’t know. I think that’s the question I’m stuck on right now.