Build nutrition agents at meal prep, not consumption — deciding what to eat while hungry is too late to override depletion

Why This Is a Rule

Nutrition decisions happen at two layers: preparation (what food is available) and consumption (what food to eat right now). Most people try to install agents at the consumption layer: "When I'm hungry, I'll choose the healthy option." But at the consumption layer, you're already depleted — hunger drives impulsive choices, the availability of easy options dominates, and willpower (already a weak intervention — see When behavior contradicts values, investigate the reward structure — not your willpower deficit) is at its lowest when metabolic need is high.

The preparation layer is the leverage point. When you decide on Sunday evening what meals to prepare for the week, you're making the decision while fed, rested, and in planning mode — optimal conditions for deliberation. The consumption-layer decision is then pre-made: the healthy meal is already prepared, the unhealthy alternatives are less available. You've converted a willpower challenge into an environmental default.

This is the same principle as Externalize high-stakes agents to tools and environment — biological memory degrades exactly when stakes are highest (externalize to environment) applied specifically to nutrition: move the decision to a point in the causal chain where your cognitive resources are best, then let the environment handle execution during the points where your resources are worst.

When This Fires

• When designing agents for eating behavior or nutrition goals
• When willpower-based approaches to eating have failed repeatedly
• When you notice food decisions degrade in quality as the day progresses or when hungry
• When applying the upstream intervention principle to health behaviors

Common Failure Mode

Relying on willpower at the moment of consumption: "I'll just choose the salad instead of the pizza." When you're standing in front of a pizza and a salad, hungry after a long day, the pizza wins because it offers more immediate metabolic reward. The time to make the salad win was Sunday, when you prepped it, placed it at eye level in the fridge, and didn't buy pizza.

The Protocol

(1) Move nutrition decisions to the preparation layer: Sunday meal prep, grocery list construction, or weekly menu planning. (2) Make the decisions while fed and rested — never plan meals while hungry. (3) Design the consumption environment for defaults: prepared healthy meals visible and accessible, less-healthy options absent or inconvenient. (4) At the consumption layer, the "decision" should already be made: "What's for lunch? The meal I prepped." No real-time deliberation needed. (5) Track preparation compliance separately from consumption: if preparation fires at 80%+ (Define agent success as 80%+ firing rate, not subjective satisfaction — felt reliability systematically inflates actual performance), consumption usually follows automatically because the environment handles it.