What should I charge for this plate? A costing app for working kitchens, taken from a zip file to a live closed beta in six days. Every decision below is dated, with what it was chosen over.
Cooks price a dish by looking at the rest of the menu. The tools that do the maths properly feel like accounting software, so nobody opens them mid-service.
Recipe costing won on three tests: least competition in the band a small kitchen can afford, closest to the product lead's own daily work in kitchens, buildable by one product lead directing AI agents.
The brief pointed at a file that didn't exist, and the one that did would have failed every build. The first day went on checking the inheritance before adding to it.
The engine was right and the surface was wrong. One prototype before any rewrite, so the week wasn't bet on an untested guess. Prototype one went too: "not a flow, buttons everywhere". Prototype two gave us the line that changed the product.
if one knew what to charge they wouldnt use larder
· Benn, 22 Oct, on prototype two. Larder was the working name.
way too much an crm, SAP app than an actual interactive food cost and ingredient calculatorBenn, 22 Oct, verbatim
The product question changed from "what does this dish cost" to "what should I charge". The headline became Charge ₪X for this plate, and the cook's current menu price became optional.
Standard text recognition got none of the priced rows right on real supplier photos. A local reader got 15 of 16. A paid AI reader was rejected on day one as too expensive, then chosen two days later once it was priced: about one to two cents a page, and no two-gigabyte image to host.
The day-one wish was native iOS and Android plus web. The comparison showed app stores remove none of the cost lines (accounts, database, hosting, legal) and add review queues and fees.
Every real kitchen word was owned or crowded. Every everyday Hebrew word read wrong, or like a small town in the south. New rule: say the name aloud in Hebrew before offering it. The product lead skipped the pause we'd advised, asked for one more round, and found the pun himself: MP, as in mise en place and as in market price.
Mise en Price is good lmao
· Benn, 24 Oct, round nine
Calculator, invoice reader, five AI helpers, four redesigned pages, Excel import, a guided first run. The product lead asked for autopilot in the morning and withdrew it within the hour: small things get decided and listed, real decisions get batched into one message. One catch on the way: a footer button that wiped a whole kitchen's data in a tap. Now demo-only.
Fake-client tests passed for days. The first live run scored 19 of 24, because a rule in our own schema made the service refuse two helpers outright. Then 22 pieces of real kitchen paperwork exposed two more gaps: iPhone photo formats, and delivery notes with no prices on them.
The deploy brief said one host. The product lead caught that it contradicted the plan from the day before and switched. An hour later, on cost, switched back. Nothing had been created on the wrong one; one agent was stopped before it touched a file.
The app was English-first from day one. Hebrew went in so the closed beta could start with the chefs and cooks in the product lead's own network. About 800 strings, estimated at four to six days. The whole chef-facing interface landed in roughly three hours. The product lead found the language switcher pointing the wrong way within a day.
why is the language button brown when it isnt the language selected
· Benn, 24 Oct, verbatim
A security pass against our own app found uploads capped by file size but not by pixels: one small image could decode to a gigabyte. Fixed before any chef arrived, with security headers and an invoices page that went from two seconds to a fifth of one.
Domain, Google, Apple and email sign-in, error monitoring, a closed-beta gate, all in one day. The first real invoice, thirteen lines photographed off a laptop screen, came back right. It felt slow while reading, so the next fix is the waiting screen, not the model.
really cool
· Benn, 26 Oct, on the first real invoice the live app read
M and P built from strips of masking tape, the kind every kitchen uses to label containers, in the app's rust colour. Three rounds of variants; the product lead picked one and it went onto the icon, the sign-in and the share image the same day. A trademark sign went on every page too, and came off within hours.
The plan had been four cities at once. Market research cut it to one market at a time, with London as the first one abroad.
So the app is ready when London comes, it now works in £, and that took more than a pound sign. UK invoices mix zero-rated food and taxed items on one page, so tax is now read line by line. Packs read as printed (1 x 5kg, 12 x 250g, punnet, case), pounds and ounces convert, and British names match: coriander isn't cilantro, prawn crackers aren't prawns. A British demo kitchen with seven dishes and GP shown next to food cost.
In October an allergen helper was cut: that's the chef's job. Getting ready for the UK changed the answer: UK kitchens have to tell guests about 14 allergens, so the app now tags them on ingredients, shows them on every dish and prints the menu sheet. The sheet says it shows what was ticked, and to check the supplier's labels.
London first was the plan for three days. The closed beta stays local, with kitchens in Israel the product lead can visit, and London is the step after the beta, with the app already working in pounds.
we are going local in israel and trying to enter london, but that will probably only happen after the beta
· Benn, 3 Nov
A closed beta with kitchens in Israel, written to one by one, free until release. After two weeks of use, each chef gets one question: what would you pay? No chef outside our own circle has used it yet, and nothing here is claimed before it's measured.
changes merged in six days, each through automated checks before merging.
unit tests, import to launch day.
Hebrew invoice rows read correctly, on our own photos.
known vulnerabilities in dependencies.
spent on about 170 real AI reads during testing.
running cost now. $0 until the first kitchen was in reach.
for the Hebrew interface, estimated at four to six days.
the invoices page under a kitchen's worth of data.
Photograph a supplier invoice. Every line is read and priced per kilo.
Every dish that uses it is recosted, with its GP and its allergens.
Then the one question it answers: what should I charge for this plate?
Or look around: misenprice.app ↗
The other app: WithOut Contact →Follow the log: RSS feed · Instagram
Back to top ↑