UNIVERSITY OF EL SEGUNDO / CLASS CHALLENGE / PROPOSED NOV 1, 2026
Put an AI in charge of buying food
for 25 miles. Then prove a person could check it.
Teams design an AI buying desk for one real kind of kitchen within 25 miles of El Segundo: a school cafeteria, an office kitchen, a pantry, a restaurant group or a household market list. The AI does the paperwork. People sign the commitments. Every decision has to explain itself in plain language.
The brief
One kitchen, one season, one desk
Pick one kind of buyer. Design the desk that would buy its food for one season from what’s actually available within 25 miles of El Segundo City Hall: the certified farmers markets, the wholesale produce district downtown, the ports, and the grocers that already serve the area. The food report is your starting map, with every market day and number sourced.
Your desk has to do four things:
- Forecast. Turn a menu or a household list into what to buy each week, and say how sure it is.
- Source. Choose where each item comes from and why: price, distance, freshness, season, and the grower’s terms.
- Move it. Plan how it gets to the kitchen, and what goes back on the empty trip (crates, compost, surplus for a pantry).
- Show its work. Publish a plain-language receipt for every order that a person could read in a minute and overrule.
Then answer the question the whole report asks: what would this look like in 2028 to 2030, through automation, transportation and the best ways of growing? And what could El Segundo start in 2027 to build toward it?
The rules
What the AI may and may not do
- The AI may draft orders, compare prices, plan routes, flag waste early, suggest substitutions within a written spec, and write the receipt.
- The AI may not, on its own, sign a contract, spend past a limit a person set, drop or add a grower, set what a grower is paid, approve any allergy or diet substitution, act on a food-safety hold, or decide who goes short when supply runs out.
- Use public or made-up data only. Practice data must be labeled as practice. No real household’s orders, no private people, no logins to anyone’s accounts.
- Label what you know. Use the site’s labels: fact from an official source, listing from the organization’s own page, estimate for your arithmetic or a model’s output.
- Say which AIs you used and what each one did, the way every page on this site does.
How it’s judged
Proposed rubric, five parts
Can a person audit it?
Pick any order at random. Could a cook, a parent or a grower understand why it was placed, from the receipt alone, in under a minute?
Is it fair to growers?
Does the desk keep small and nearby growers in the running, pay on stated terms, and avoid quietly moving everything to one vendor?
Does less get wasted?
Does the plan forecast spoilage, route surplus to recovery, and use the return trip?
Does it hold up when things go wrong?
A late truck, a heat wave, a recall, a price spike. Show what the desk does, and where it stops and asks a person.
Is the 2028 to 2030 picture honest?
Speculation is welcome when it’s labeled. The best entries separate what exists now from what they’re betting on.
Four weeks
A suggested pace
- Week 1, the map. Choose your buyer. Map where its food could come from inside 25 miles, using the report’s sources.
- Week 2, the desk. Build the first version: forecast, sourcing, route, receipt. Practice data only.
- Week 3, the stress test. Run the bad week: a recall, a shortage, a late delivery. Tighten where a person has to step in.
- Week 4, the horizon. Present the desk, its receipts, and your 2028 to 2030 version with three steps the area could start in 2027.
What the AIs said
We asked several AIs the same question
Before writing this challenge, we asked several AI models how they would run food buying for the whole 25 miles, how they’d redo distribution, what 2028 to 2030 looks like, and what should stay human. Three of them landed on the same last line without seeing each other’s answers: deciding who eats when there isn’t enough stays with an accountable person. Their answers, labeled as AI speculation, are on the food report. Treat them as prompts to argue with, not as facts.
Where this goes next
From a class to the area
If a team’s desk holds up, the next step is a small, real pilot with willing kitchens, run in the open. Industry Next, Michael Hoydich’s independent studio, has proposed helping run that kind of pilot. That is an offer, not an arrangement with anyone yet.
About this page
How it was made
Drafted by Claude (Anthropic), November 1, 2026, America/Los_Angeles, at Michael Hoydich’s request. The challenge, rules and rubric are a proposal. Facts about local food live on the food report, where each one opens to its source. See something wrong? Send it through Share an update.