Three right diagnoses, three times worse

We have a tool, Lens, that opens our pages in a real browser and checks they look right. It took screenshots.

Then came a bug it could not see: in our own chat the view jumped by itself, with nobody touching anything.

A picture cannot show a movement. So we read the code instead, found an explanation that fit three times — and made it worse all three times. We didn't need more hours. We needed an instrument.

What we built

Lens can now record. But it isn't the video that finds the bug. It's a measurement: did the page move without anyone touching it? A jolt after a click is just a page scrolling. A jolt without one is a bug. Only afterwards are the frames cut out — to look at, not to search in.

We deliberately did not hand 600 frames to an AI and ask if it saw anything. A model asked to find something will find it — including when there is nothing. The most important property is that it can say no.

What it gave us

Five hours of reading code gave three wrong answers. A hundred seconds of recording found the bug on the first try — plus one we hadn't seen.

It does not record your screen; it runs its own browser.