Nobody asked for the AI feature

"The investors love it," the founder said, and that sentence is how a quarter of the roadmap got spent. The feature was an AI assistant bolted into the corner of the product, a chat box that could answer questions about your own data in natural language. It demoed like a dream. It showed beautifully in the raise. And in the months after it shipped, the usage logs told a flat, embarrassing story: almost nobody opened it twice.

Meanwhile, the single most requested item in the support queue was a button that exported a table to a spreadsheet.

You can feel the gravity that pulls a team toward the assistant and away from the button. The assistant is exciting. It is the thing you say first when someone asks what's new. It makes the founder sound like they are building the future, and on a fundraising call that is not a small thing. The export button makes you sound like you are building a slightly better version of software that already exists. One of these gets applause in the room. The other gets used at 8 a.m. by someone with a deadline.

You can build for the room you pitch in or the desk it ends up on. They want different things.

The users were not being unimaginative. They were being honest about their day. They had a workflow that already worked, mostly, and the one place it broke was getting their data out so they could do the last mile somewhere else. The AI assistant did not touch that pain. It offered them a new and impressive way to ask questions they mostly already knew the answers to. So they admired it once, the way you admire a concept car, and went back to copy-pasting cells like animals because the export button still didn't exist.

The trap here is not that AI features are bad. Plenty earn their place. The trap is using "the investors love it" as a substitute for "the users need it," because those are two different validations and only one of them renews the subscription. Investor enthusiasm tells you the story is good. It tells you nothing about whether the thing gets opened on a Tuesday. The roadmap that confuses the two ends up full of features that are wonderful to announce and quiet to own.

There is a cheap test that cuts through it. Before building the impressive thing, look at what people are already trying to do in your product and failing. The friction is right there in the support tickets, in the rage-clicks, in the workarounds people have invented. That is demand with the marketing stripped off. An AI feature that removes one of those frictions is worth ten that merely decorate the homepage.

The assistant, to be fair, was not a waste in the raise. It did its job in the room. The mistake was believing the room's reaction was the same as the market's, and pricing a whole quarter of engineering against the wrong audience's applause.

TensorLabs has talked more than one team out of the impressive feature and into the boring one. The export button shipped. The renewals noticed. The board, eventually, did too.