#9 Restaurants Pay Rent Too

4

min read

French truffles in every risotto is waste. So is running the best model on every task.

Every time an AI model reads input and generates output, it consumes tokens. Bigger models cost more per token. More steps and more verification both mean more tokens. Tokens are the AI agent's ingredient cost — and like any restaurant, the question isn't how much you spend. It's whether you got results worth what you spent.
Running the top model on every task is like putting French truffles in every risotto. The quality might be marginally higher, but the cost-to-outcome ratio doesn't hold. Using the best model for simple data cleanup is waste — like assigning a master chef to peel potatoes. Save it for the complex analysis, the final report. That's where the best ingredients belong.
Restaurant economics and AI economics run on the same logic.


Just Staying Open Costs Money

Whether guests show up or not, money goes out every month. Rent. Staff salaries. Ingredients. Electricity. Equipment maintenance. More guests mean more ingredients and more staff. Fewer guests mean less, but the fixed costs stay.
The core of running a restaurant is simple: money in has to exceed money out. If it doesn't, the restaurant closes. No matter how good the food is.


Top-Shelf Isn't Always the Answer

Many restaurants go wrong by trying to make everything premium. Only the finest ingredients — French truffles, A5 Wagyu, direct-import Italian olive oil. Only the most experienced chefs. Every prep cook formally trained. German porcelain plates, silver utensils.
The problem is that all of this is cost. French truffles are five times the price of domestic ones. But can the guest taste the difference in a risotto? Probably not. Five prep cooks make things comfortable, but if one person can handle onions and herbs just fine, five salaries is a waste. Ten stages of quality checks produce perfection, but if five stages get you nearly the same quality, the extra five are wasted effort.
Premium isn't always right. The right answer is spending enough to deliver the quality guests notice — and not a dollar more.


When Costs Exceed Revenue

When operating expenses start exceeding income, options narrow. Reduce headcount — five prep cooks become three, one person takes multiple roles. Reduce verification — ten quality checks become five, you check only at critical stages. Adjust ingredient grade — A3 Wagyu instead of A5, domestic truffles instead of French. Cut costs where guests barely notice the difference.
All three follow the same principle: maintain the quality guests experience while cutting costs in places they can't see.


The Same Math, Different Currency

Restaurant

AI agent

Principle

5 prep cooks to 3, each handles more

5 subagents to 3, each handles more tasks

Reduce headcount

10 quality checks to 5

10 verifications to 5 critical ones

Reduce check frequency

A5 Wagyu to A3

Top-tier model to a suitable model

Downgrade where the quality gap is small

$50 ingredients → $150 revenue (3x)

Token cost vs. work output (ROI)

Judge by results per dollar spent

A restaurant spends $50 on ingredients for a course and sells it for $150. Revenue is three times the ingredient cost. That ratio shows the restaurant's health. Spend $100 on ingredients and guest satisfaction stays the same? Double the cost, same result — bad investment. Spend $30, and guests say "this place is incredible" and come back with friends? Cost went down, revenue went up — good investment.
AI agents work the same way. Restaurants measure ingredient cost against guest satisfaction. AI measures token cost against work output. Both ask the same question: not "how much did you spend" but "did you get results worth what you spent?"
Restaurants pay rent. AI agents consume tokens. Both survive on the same math.