Devraj Anand built the boring tool nobody wanted — until everybody needed it

When Devraj Anand pitched Tenscroll — a platform for evaluating whether your AI prompt changes actually helped — investors kept asking the same thing: "Isn't this just a notebook?"

"Yeah," he'd say. "It's just a notebook. The same way Datadog is just console.log."

Boring on purpose

Devraj had spent two years as an ML engineer watching teams ship prompt changes on vibes. "We'd change a prompt, eyeball ten outputs, and ship it. Then we'd be shocked when quality regressed. We had no idea if we were improving or just moving the failure around."

The problem was obvious to anyone doing the work and invisible to everyone else — which is precisely why there was no competition. Devraj saw that as the opportunity, not the warning. "If the problem is real and recurring and nobody's built the good version, that's not a red flag. That's an opening. The flashy spaces are crowded because they're flashy."

The discipline of not pivoting

The hard part wasn't the tech — it was staying on the boring problem while louder, sexier ideas beckoned every week. "The whole first year, some new model would drop and I'd feel this pull to chase it. The thing that saved us was talking to users. Every single one had the eval problem. None of them cared about the shiny thing I was tempted to build."

The one insight worth stealing

The unsexy problem with no competition is a feature, not a bug — as long as the pain is real and recurring. The crowd is in the flashy space because it's flashy.

Devraj's edge wasn't vision. It was the discipline to keep building the thing people needed instead of the thing that would trend. In AI right now, that discipline is the rarest asset there is.