Invisible AI beat visible AI, and the numbers were not close

The same capability failed as an AI feature and succeeded as a button. Neville Medhora, who has roughly 100,000 tagged subscribers on Kit, was in the beta group for its MCP connector and stopped using it inside two weeks. Kit then shipped the one genuinely useful thing the connector could do as an ordinary product feature, without calling it AI. That version worked.

Why the connector failed

Not capability. Not accuracy. He could not think of anything to ask it.

"I was just like: I don't know what to ask it. This is the biggest problem. I have access to all the information and it plugs into an AI, but what do I do with it? And that was the biggest problem."

And when he did ask, the existing interface was faster: "It's actually easier to just click the report than to ask ChatGPT, then it talks to the MCP, and then it finally comes back."

He reports the whole beta group had the same experience — people pulling reports they could already pull.

The one question that broke the pattern

Someone in the beta asked it to find the most influential people on the subscriber list. That required cross-referencing roughly 100,000 email addresses against social platforms and tallying followings — work a dashboard cannot do and a person would never do by hand.

"The top 20 people had million-plus followers. I had no idea these people even followed me."

What the vendor did next

"So what Kit did was says, you know what, screw all this MCP stuff. We got really excited about it. Everyone did… No one even knows that you can do all this stuff in the back end and it would take a really long time. So let's just do it for them."

And the detail that makes it a product lesson rather than an anecdote:

"They don't call it that. They don't say it was gathered with AI. It just says: here's the influential people."

"So now it is a more valuable piece of software. And they don't have to confuse everyone with all this AI stuff."

The rule this produces

A blank prompt attached to a customer's data transfers your hardest problem to them: knowing what is worth asking. Most customers cannot do it, will not admit it, and will churn quietly at around the two-week mark.

Ship the question, not the interface. Then stop advertising which technology answered it.

Artificial Reality note: this account of Kit's product decisions is Medhora's first-hand experience as a beta participant and customer. We have not independently verified Kit's internal roadmap reasoning, and Medhora is himself unsure of the resulting feature's exact name.

AI tools named in this report

Tools named by the operator on the record. Inclusion is reporting, not endorsement.

Where this comes from

S2E1: AI should have killed his career. He's making $1M a year instead — the full interview with Neville Medhora. Listen or watch: YouTube, Spotify or Apple Podcasts.