You don't need a smarter model to make food cheaper. You need to delete a middleman.
That's the whole argument. Everything below is just the receipts.
Open any grocery or delivery app. The first three results aren't the closest store, or the cheapest one. They're the ones who paid to be there.
That's not an accusation, it's the business model. Delivery platforms commonly take 15–30% commission off every order. Sponsored placement is its own line item on top of that. None of it comes out of the platform's pocket, it comes out of the price you pay, baked quietly into the menu before you ever see it.
Multiply that across every grocery store, every restaurant, every small food business that wants to be findable online, and you get a strange situation: the discovery layer, the part that just tells you a store exists and what it sells, has become one of the most expensive parts of the entire food supply chain. Not farming. Not logistics. Being found.
Take that tax out, and prices don't need a subsidy to drop. They just drop, because the markup that was funding it isn't there to fund anymore.
We think that number is north of 30% for a meaningful slice of the market. Call it a working thesis, not a peer-reviewed study, but it's the kind of thesis you can actually go test, because the fix is small enough to build in a weekend.
Here's the uncomfortable part: the technology to remove the discovery tax isn't new, exotic, or AI-shaped. It's boring. It's:
That's it. That's the whole disruption. No model, no inference cost, no API bill. A map and a catalog, both open, both free to read and free to write.
This is exactly what we open-sourced last week with the Food Freedom Map, a pin-drop map built on Nostr where any store, farmer, or restaurant can list themselves and their catalog, and any client can read that data, forever, without paying anyone for placement. No ad auction. No "boost your listing" button. The map doesn't rank you by budget, because there's no budget to rank by.
The interesting question isn't whether this can work, the protocol's already running. It's how fast you could migrate an entire region's worth of existing stores onto it.
Say you wanted to bootstrap a city overnight. The shape of it looks like this:
Step 1 — Pull the data that already exists.
Most of what you need is already public. OpenStreetMap's Overpass API will hand you structured shop/restaurant/category data for an entire city in one query, no scraping required. Government open-data portals (business registries, food-safety permits, market licenses) fill in the gaps OSM misses. You could also scrape something like Google Maps for the same fields — plenty of projects have — but that's walking into a Terms of Service fight with a company that actively fingerprints and blocks exactly that kind of traffic. Open data gets you 80% of the way there with none of the legal risk. Start there.
Step 2 — Normalize it into one schema.
Name, location, category, that's the minimum viable pin. This is the boring-but-essential part: every source names things differently, so you write one small mapping layer and you're done. Genuinely a script, not a system.
Step 3 — Drop it on the open map as unclaimed pins.
The store exists on the map whether or not the owner has ever heard of your project. That's the point — discovery doesn't wait for adoption.
Step 4 — Give every pin a free, editable post.
This is the part that makes it stick. The owner finds their pin, claims it, and can edit their catalog directly — what they sell, what they're asking for, what they're giving away — for free, forever, the same way anyone claims a Google Business Profile today. Except there's no algorithm deciding who shows up first, and no fee for not disappearing.
Run that sequence and a city's worth of stores can go from "invisible unless they pay" to "on the map, editable, free" in roughly the time it takes to write and run an import script. That's not a five-year roadmap. That's a weekend project sitting on top of a protocol that already exists.
The headline isn't anti-AI for the sake of being contrarian. It's pointing at something specific: this particular problem doesn't need a smarter answer, it needs a different incentive structure. No amount of better recommendation algorithms fixes a market where the algorithm's owner gets paid by the seller, not the buyer. You can't optimize your way out of a conflict of interest. You have to remove it.
That removal is mechanical, not magical. It's a public dataset, an open protocol, and a free listing. The kind of thing one engineer can ship, not the kind of thing that needs a foundation model behind it.
The repo's open, the protocol's documented, and the gap between "idea" and "your city is on the map" is exactly as small as it sounds. If you want to take a swing at the import-script side of this — OSM importer, open-data normalizer, claim-flow UX — that's a wide-open, genuinely useful first contribution.
👉 github.com/Djowda/food-freedom-map Fork it. Point it at your city. Tell us what broke.