I have been on blood pressure medication for over a decade. At some point I fell off of strictly watching sodium intake as well as I should (it happens), and I decided to get back on the horse recently. My biggest foes have always been: flavor and freshness. Flavor goes away without salt, and usually you make up for it with fresh foods, which puts you at the store all the time, or falling back to bad snacks otherwise.

So, I sat down with my Genie and started what turned out to be a multi-day dozens of iterations-long conversation about building heat-and-eat bowls that only used shelf-stable or frozen ingredients. I got 5 different flavor profiles out of this discussion and set about making the first.

I told the AI I needed a week’s worth of meals out of one pot, and so it was adamant that its creation was 7 ~1lb servings. I asked it to check its own math before I got started on the cooking. But when I got done dividing it, I had a serving left over! Concerned, I went to a good, old-fashioned calorie counter app and put in all my ingredients. Low and behold, dividing that recipe by 7 was going to put me well beyond my desired meal size and calorie count. Luckily for me, the fix was an easy move from 7 to 8 servings, and the overall macros stayed in line with my desired outcomes. But using only AI to validate AI could have been real problem here.

The second batch I made, it way-way undersold the amount of liquid I needed for the grain (quinoa in the first, rice in the second, which really needed 1.5x liquid). That led to undercooked chicken and a general triaging of that whole batch of food. I had to go look at outside sources for recipes here to make sure I figured out the new ratio correctly. A second set of eyes (even AI eyes) probably would have been valuable here.

While I eventually got to ground truth (and some tasty meals, at that!) I probably should have worked out my validation criteria before I launched, as well as my review process, and I definitely should have stepped away from the AI back to the land of cold, hard facts before I started, too. I also probably should have started cooking for tomorrow after I ate, but that’s another story.

I know these lessons from development:

• When you work out your validation criteria first, your system will always match those (for better or for worse). When you don’t, you end up back-porting criteria on later, and it’s much more of a pain.

• When you gather your prerequisites at the outset, you won’t be surprised by them. But when wait, new requirements (like additional liquid) can really throw you for a loop.

Sometimes I’m bad at bringing those lessons back to the rest of my life.