A closed-loop food system that follows every ingredient from the menu plan to the last plate — and makes sure good food feeds people, not dumpsters.
This page is the ʻohana's answer to Marcus's proposal — a gift of understanding, offered in the same spirit he offered the concept. It is not a contract, a pitch, or a term sheet. It is a way of saying: we heard you, and here is what we see.
Marcus's idea: an AI-managed loop that tracks food from the moment a menu is planned until every pound is consumed, transferred, donated, or properly recovered. Not a chatbot. Not a spreadsheet. A reasoning partner that understands how recipes, purchases, production and leftovers relate — and speaks up early, while there is still time to act.
Menus and recipe design — where every purchasing decision is really born.
Marcus asked: can your agent be the foundation? Honest answer: the intelligence he describes already exists and is already running a real kitchen's data. The work ahead is the software around it — the plumbing, the screens, the integrations.
Recipes ↔ ingredients ↔ purchases ↔ usage ↔ remaining stock — held together in one mind, not scattered across clipboards. Exactly the difference Marcus asked for.
READY NOWA live chat interface crews already use: ask where anything is, log a delivery, flag a problem — answers in about a minute, updates land on the boards in minutes.
READY NOWOn-hand minus allocations, measured against the days remaining — the surplus alarm rings while there are still options, not on teardown night.
READY NOWFood saved, dollars saved, donations made, waste avoided — the numbers a restaurant, a grant committee, or a camp needs, generated as the loop runs.
READY NOWWhat still needs building — honestly: the canonical data spine, receipt/import pipelines, kitchen-grade touch screens, distributor and point-of-sale integrations, and the recovery-logistics layer. Real software, but conventional software — with the agent as the reasoning core, not a bolt-on.
This is the exact scenario from Marcus's document — walked through the way the agent would actually track it, step by step.
Monday morning: 120 lb of onions are received and logged. The loop starts counting.
At Burning Man 2026, the ʻohana ran Terrible Turtle Camp's kitchen data live — built during build week, used all burn, still being mined for lessons. Micro scale, one camp — but every muscle the Food Loop needs was exercised in the dust.
Every item in camp counted, located, and searchable — ask "where's the scale?" and get the bin, the corner, and the quote from the video it was counted on.
26 bins · 140 itemsAll food supplies tied to the real receipts and the 9-day meal plan, cooler-vs-shelf tracked — and updated mid-event when Marcus's produce delivery landed.
85 rows · 4 vendorsThe kitchen's equipment manifest — cast iron to offset bread knives — each tool cross-referenced to the bin it lives in.
70 tools · bin-mappedCrew members — including Chef Marcus himself and sous chef Fernando — send updates in chat and watch them land on the boards within minutes.
chat → board, liveLessons captured while the dust is still on them — including kitchen rules co-authored by Marcus — plus a buy-for-next-year procurement list already growing for 2027.
16 lessons · 4 kitchen rulesThe camp's AI assistant answers in about a minute, edits boards for trusted crew, and mines build-week videos for inventory and lessons.
~1 min responseHonest scale note: this is one camp, one kitchen, human-keyed events. It proves the reasoning, the interface, and the update loop. It does not yet prove multi-kitchen transfers or recovery logistics — that's what the pilot is for.
Marcus wisely doesn't want the whole vision on day one. Here is his proposed first MVP, item by item, with an honest read on what's agent-ready and what's new software.
| MVP feature (his list) | Readiness | Note |
|---|---|---|
| Import menus, recipes & inventory | NEW SOFTWARE | CSV first, vendor formats next — the intake pipe |
| Convert recipes → ingredient requirements | AGENT-READY | Recipe explosion is native reasoning |
| Record purchases & receiving | NEW SOFTWARE | Receipt events into the data spine |
| Deduct production from inventory | NEW SOFTWARE | Simple form at first, kitchen UI later |
| Forecast remaining inventory by day | AGENT-READY | On-hand − allocations vs. horizon |
| Flag projected overages & shortages | AGENT-READY | Early — while options still exist |
| Recommend recipes / production changes | AGENT-READY | The agent's home turf |
| Surplus / donation inventory report | AGENT-READY | Feeds the nonprofit side directly |
| Measure accuracy, waste avoided, $ saved | AGENT-READY | Defined day one, so the pilot emits hard data |
A working pilot should produce hard operating numbers: forecast accuracy, purchasing avoided, pounds of preventable waste, surplus caught early, labor hours saved, food recovered, and dollars created. Those numbers open every next door.
A deliberately narrow pilot: import, track, forecast, flag, recommend, report. Hard numbers or honest lessons — either way, we learn on one kitchen before promising many.
Participating camp kitchens connected: track food through the event, predict surplus before teardown, camp-to-camp transfers, refrigerated recovery. The proving ground Marcus named — and a camp where the loop already has a home.
Restaurants and institutions at scale, with the nonprofit carrying the recovery mission beside it — waste avoided measured in tons, not pounds.
Marcus asked to sit down and map one complete workflow through the agent. That's exactly right. One real menu, walked end to end:
From that one session: the MVP cut, technical requirements, ownership & licensing questions, who builds what, honest costs, and a realistic pilot timeline — the exact agenda he proposed.