I started LADLE because I could no longer defend, to myself, paying twenty dollars a month to a company whose whole plan for that money was to become a larger company.
The math on AI subscriptions in 2026 is a strange thing. The frontier models — Claude Sonnet, GPT-5, Gemini 2.5 — cost more than a hobbyist could ever afford to serve, and yet the consumer product built on top of them costs the same twenty dollars a month that a Spotify subscription costs. The reason is competitive: every provider is racing to acquire subscribers before the pricing settles, and none of them can charge what it costs. So they charge twenty, absorb the loss on the heaviest users, and invest what remains from the middle 80% into ads and moats and next year’s lab. That is not a criticism. It is a description. It is what companies at their stage have to do to survive.
But somewhere in the middle of that middle 80% is a real number that could just as easily do something else. If a company committed, structurally, that eight of those twenty dollars would leave the building forever — go to a specific, verifiable program that fed people who were going to be hungry that month — the product would still exist. It would just be a different kind of product. Not a philanthropic product. A product that had one operating decision baked into it that most other products in the same category couldn’t make.
That’s what LADLE is. Same class of model, same twenty dollars, one different operating decision. The mechanics: eight dollars of every subscription becomes ten meals a month through the United Nations World Food Programme via ShareTheMeal, at eighty cents per meal, transferred on the last day of every calendar month with a reference number I publish on the same day it clears.
I want to be honest about what this is not. It is not a solution to any of the things it’s adjacent to. It doesn’t fix world hunger; it funds a specific program that already works and could use more money. It doesn’t make LADLE the “ethical” AI subscription; that word does more damage than good and I don’t want it applied to me. It doesn’t offset the environmental cost of AI inference; the money funds food, not carbon, and pretending otherwise would be a different kind of dishonest. It doesn’t make you a better person for subscribing. It just means the money you were going to spend on a specific product does an additional specific thing while you use the product.
The commitment has to be structural, not optional. If I offered a “donate a portion of your subscription” toggle, the average opt-in rate for these things is under ten percent, and the whole point of the design is that the meals are unconditional on subscriber generosity. Making it a checkbox would make it marketing. Making it a fixed portion of the base price makes it operating cost. I can’t offer a Pro tier that opts out. I can’t offer a retention discount on cancel — offering a discount would offer to fund fewer meals. I can’t raise the price. I can’t lower the price. Every one of those doors has been welded shut before the company started, and that’s the design.
What this leaves is a very unusual small business. The margins are permanently constrained. The growth arithmetic is boring. The runway comes from other people’s payroll, which is to say, from being small and staying small until the compounding math on model inference prices catches up to us. That’s the play. No hockey stick, no exit narrative, just a subscription business whose most distinctive line item is a monthly transfer receipt.
The reason I’m confident this is worth building: because the alternative was not building it, and I’ve tried the alternative. Every AI subscription I’ve paid over the last three years has been perfectly good, and none of them have made me feel like the money did anything I could point to at the end of the month. That’s a real, small, boring feeling. Enough of them together are the reason people cancel things. Not because the product is bad, but because the product is fungible. LADLE is not less fungible on the product side — it’s a chat window with a frontier model. It’s a little less fungible on the receipt side. Sometimes that’s enough.
If you subscribe, I’d ask two things of you. First: cancel us the day we can’t show you the receipt. The whole trust mechanism collapses without that, and I’d rather you leave than pretend nothing happened. Second: if the meal math is what got you to sign up, please also read /impact/tax — because you cannot claim this on your taxes, and if a tax deduction is what you wanted, ShareTheMeal takes donations directly and the money goes to the same place more efficiently. I’d rather you donate there and use whatever AI subscription you’d otherwise pick. This company only makes sense for people whose AI spend is going to happen anyway.
The receipts are on /impact/reports. The numbers behind the company are on /open. Everything else is on the rest of the site. Thank you for reading a long letter.
— The founder, LADLE PBC