On Wednesdays, I want dinner to feel composed but not demanding.
By midweek, my tolerance for complexity decreases. I still want something balanced and thoughtful, but I do not want to build layers from scratch or stand over a pan reducing anything. I want structure without ceremony.
On this particular Wednesday, I had a large skin-on salmon filet. Salmon has become one of my most reliable canvases. Rich enough to carry texture. Forgiving enough to roast without drama. My pantry usually holds enough nuts, seeds, mustards, oils, and vinegars to create contrast without creating work.
I was not trying to invent something new. I was trying to elevate something ordinary.
So I asked ChatGPT:
↳ The Aided Chef:
I have a large salmon filet. I want something high-impact but simple for a Wednesday night. What’s a way to add texture without a lot of prep?
Among the several options offered, a seeded crunch top stood out. Olive oil. Dijon. Sesame seeds, pepitas, chopped almonds. Roast. Finish bright.
It felt right for the night.
I rubbed the salmon with olive oil and Dijon, pressed on roasted and salted pepitas, toasted sesame seeds, and raw chopped almonds, and added black pepper. I didn’t salt the fish itself.
That choice was deliberate. The pepitas were already salted, and I assumed that would be enough. I didn’t want the dish to tip into sharp or overly salty territory. With mustard and sesame already in play, I was thinking about restraint, not structure.
Everything went into a 425°F oven. Around the 12-minute mark, I pulled the salmon out and checked the center. It looked underdone — darker and softer than I wanted — so I returned it to the oven. I was calibrating visually, not structurally. In the end, it roasted for about 18 minutes.
When I finally pulled it from the oven, the seeds had darkened and the surface had tightened in a way that looked deliberate. The fish felt set. I took that as a sign that it was done.
The salmon itself was moist.
But the crust felt dry. The flavor, while pleasant, was muted. The seeds sat on top rather than integrating with the fish.
My first instinct was simple: I must not have used enough salt. Or maybe I left it in too long. Instead of asking for a different recipe, I went back to ChatGPT and reframed the problem:
↳ The Aided Chef:
I made the seeded salmon. The fish was moist but the crust felt very dry and the flavor was a bit bland.
I used roasted salted pepitas, toasted sesame seeds, and raw chopped almonds. I rubbed olive oil and Dijon on the fish, did not salt the flesh itself, and roasted at 425°F for about 18 minutes.
What likely caused this, and what would you adjust next time?
That second prompt was different in one specific way: I described the outcome before asking the question. I listed my variables. I named what I observed rather than what I wanted fixed. That framing changed what came back.
Not “fix this.”
Not “give me a better recipe.”
But: here is what happened. Here is what I did. Help me understand the structure.
What came back was not a new recipe. It was analysis.
↳ ChatGPT:
Salt on toppings won’t season the protein beneath them during a short roast. Surface salt stays surface.
You layered mostly dry elements at high heat. Without a stronger fat base, the crust firms instead of binding.
Eighteen minutes at 425°F can exaggerate surface dryness even if the interior is just done.
A final gloss of acid or oil can restore surface moisture and sharpen salt perception.
This is what I took from that response:
I wasn’t under-seasoning. I was misplacing salt.
I wasn’t just cooking a few minutes too long. I was amplifying imbalance.
From that exchange, a few principles began to clarify. Not rules. Just patterns I could see more clearly.
Salt, I’m beginning to see, is structural.
I was focused on avoiding excess. What I missed was placement. Salt on a topping does not replace salt in the protein itself. Where it lands determines how it functions.
That was not a seasoning problem. It was a placement problem.
Texture is cumulative.
Dry layered on dry at high heat does not produce integration. It produces rigidity.
I had built contrast. Not cohesion.
Heat magnifies small imbalances.
Time does not simply finish cooking. It intensifies whatever structure is already in place.
I thought a few more minutes would make it safer. They made it firmer.
And finishing is calibration.
The last touch is not decorative. It is corrective.
It restores moisture. It sharpens perception. It gives the dish one final chance to rebalance.
If I make this again, I won’t change the idea, I’ll change the structure. I’ll salt the fish itself and let it rest. I’ll build a base that holds rather than thins under heat. I’ll pull it from the oven a little earlier than feels safe. And I’ll finish it with intention, not as garnish, but as an integral part of the dish.
The concept was sound. The execution faltered in small, cumulative ways. I didn’t need a new idea, I just needed to understand what had happened. AI helped with that.
On Wednesdays, I want dinner to feel composed but not demanding. This salmon did not deliver that, but it gave me something more useful. That is a reasonable trade for a Wednesday.
If you have ever used AI to troubleshoot something that did not work, I would genuinely love to hear how you framed the problem.
Did you describe the outcome in detail? Did you list your variables? Or did you ask for something entirely new?
I am always curious how other people think through that moment when something is almost right but not quite there. Share a story like that in the comments. Not the recipe. The diagnosis.
