The Chile and Lime Question

Last week, I had an hour to kill while I waited for my son’s soccer practice to end, so I passed the time reading Vegetables Unleashed, a cookbook by famed chef and humanitarian José Andrés. This passage stood out:

“I love cookbooks—I have thousands in my library and spend countless hours losing myself in their pages. [...] But even so, I recognize that cookbooks have limits—they alone can’t guide you to those special, transcendent food moments. Eating corn out of the field, husked and devoured right there, or plucking a tomato from the plant and slicing into it, so it sheds warm tomato tears onto your cutting board-those are moments that only you can create.”

He was making the point that the knowledge one could gain from books is insufficient for creating something meaningful. That knowledge must be paired with what we learn from first-hand experience.

Part of the reason is that plenty of knowledge exists outside “official” sources. For example, Andrés experienced a surprising chopped-fruit salad in the street markets of Mexico. “Nearly every culture in the Western world has figured out that fruit and cheese belong together,” he writes. But the street vendors, working outside of those official sources, were “the only ones smart enough to invite chile and lime to the party.”

This week, I was also sketching ideas for my organization’s AI use policy. 

One thing was obvious as I did so: most of the interesting work would be landing on the right principles. It was clear that with such a fast-moving technology—and societal views on it—there would be no way to articulate everything as a series of rules that eliminated all the gray areas. Instead, we’d need people to make values-aligned decisions when they found themselves in those gray areas.

But I soon realized that even once we landed on the principles, what I’d want to say to my teammates in their development meetings would be much different and much sharper.

For example, an organizational principle might be, “You’re responsible for your work. If you turn in work that's sloppy, that’s on you, not the AI.” But when talking to someone about their development, I’d want to say, “Look, if you’re just sending me the first AI result, you’re not that useful to me. I can get that on my own.”

That reaction partly came from talking with a colleague about our shared frustration with people sending us substandard AI work, even though they could have produced better results if they'd done it from scratch. They had taken off their critical thinking hats. 

Saying that out loud, I realized the principle itself is almost a “well, duh.” Of course you shouldn’t turn in bad work—that’s not a bar anyone should be proud of clearing. But if I actually care about someone’s development, getting to the baseline isn’t the core conversation. They need to hear where they have to push to stay valuable.

Another version of that sharper development point: “If you don’t know anything different than the AI does, you’re going to have a hard time going forward.”

That thought reminded me of José Andrés’s point about cookbooks—that reading them is insufficient to doing meaningful work. This is especially true in a world where the thousands he’s read put him at an expert level, but still fall short of the number of cookbooks AI has access to. The chef is most relevant when they pair that general knowledge with “local” knowledge—for example, how the specific vegetable they’re working with differs from the norm and calls for a different preparation, or knowing one’s guests’ tastes well enough to build a meal around their preferences.

In our school setting, that local knowledge might mean knowing students and families well enough to make better sense of, and better use of, AI-powered analysis than the next person. In a commercial setting, it might mean having talked to enough customers that you can take the generic copy AI produces and translate it into terms that resonate most.

In a world where much of the general knowledge will be accessible to everyone, the question I’d want to put to everyone on my team is, “Are you working in a way that gives you access to the chile-and-lime-style local knowledge?” And my instinct is that I wouldn’t be doing them any favors by asking a softer version of that question or appealing to generically good principles for using AI. The sharpest version is likely what’s needed to spur action on their development—even if it causes anxiety—because it would be a mistake to discover too late that they’d built the wrong knowledge and the wrong skills.