How to Tell If a Product Is Really AI-Native
A piece by Duy Cao came through the TLDR Fintech newsletter at the end of July with a single question for telling apart a product built on AI from a product with AI features attached to it: if the AI disappeared tomorrow, would the thing still work? Cao is writing to founders, in an article called What Fintech Founders Get Wrong About AI-Native Development, but the question is just as good from the buying side, and it is short enough to run on any app you are considering in about a minute.
He credits it to CRV, an early-stage venture firm that invests in the category it is grading: "The simplest way to know if a company is AI-native is to ask one question: if you remove the AI, does the product cease to function?" CRV sorts three tiers with it. AI-native products were built on AI from the ground up. AI-first products moved AI to the center later. AI-enabled products are "enhanced with AI features, where the core value proposition existed before AI was added and would survive without it."
Cao's version is pointed at founders: "If someone removed the AI layer from your product tomorrow, what would actually break?" I build a budgeting app, so I answered it about Trupocket.
Running it on my own product
Trupocket connects to AI assistants through an MCP server. Model Context Protocol is an open standard that lets an assistant like Claude work with an outside service, and here the outside service is your ledger. With it gone, what remains is a budgeting app with manual transaction entry, categories, budgets, scheduled transactions, mortgage tracking, and reports, sitting on a REST API with 60+ endpoints. An API in this sense is a documented set of web addresses any program can call to read or change the same data the web app shows you. Everyone on the free plan would see no difference at all, because assistant access starts on the $2.99 Premium tier.
So almost nothing breaks. By CRV's scale, Trupocket is AI-enabled, and Cao frames that grade in a way I agree with: "If the honest answer is 'most of the product still works fine,' you're AI-enabled, and that's a legitimate thing to be, as long as you say so."
There is a second question worth asking any app, which is whether the AI showed up before or after the thing it operates. Trupocket's dates are public. The REST API launched on December 7, 2025. The web app followed on February 11, 2026. The MCP server arrived on June 11, 2026, about six months later, and it runs inside that same service under the same sign-in and the same permissions your own scripts use, which is the access model I wrote about in what makes it safe to let AI touch your money.
What would have to change
The feature I most want an assistant to own end to end is receipt scanning, where you photograph a receipt and the transaction writes itself. It sits in the On the Horizon section of the roadmap, the furthest out bucket there is. Bank syncing sits in the same bucket, which is why entry is manual today, and manual entry is exactly the sort of chore a product designed on AI from the start would have removed before it shipped. Both would move the grade. Neither has shipped, and I would rather write that down here than let a label on a pricing page imply otherwise.
How to run this on an app you are considering
I have spent this series on the parts of a money app a customer can check without trusting the marketing, starting with the argument that a money app should not be built to keep you, then looking at the parts of a company you can actually verify. How a product labels its own architecture belongs on that list.
If you are weighing a finance app, the whole test is picturing its AI features gone and asking what you would still be paying for. A real product left standing means AI-enabled, which can be the right thing to buy as long as the company says so out loud. Trupocket is AI-enabled, and if that kind of plain answer is useful to you, Trupocket has a free plan.