Coding agents are awesome. You write your prompt and watch the magic happen. Many times, you can even skip the "watching" part and just do something else while the code updates.
However, there is still some friction on how you prompt the agent, especially when talking about UI/UX bugs.
I see people taking screenshots, drawing on them, then uploading those to the AI and writing a prompt along with it. It gets harder when you need to explain the user's behavior that led to an error. You have to take multiple screenshots, write down what happened, what you expected, sometimes you don't have time to take a screenshot if the UI changes fast, etc.
And after all that work, often the agent still doesn't understand the issue. It's just frustrating.
The good news is, it doesn't have to be like that! For years, we have had a way to explain and show bugs easily: screen recordings.
It's so easy to just record your screen, explain what's wrong, and send it to a colleague. Why for AI it's not the same?
At Nitpicks, we are tackling that friction. Without leaving your product page, click a button, record your screen showing a bug, and see the fix flow to your GitHub repository automatically on a pull request.
Top comments (1)
Recording catches the behaviour, which a still can't do. The gap is the same one though: the agent gets pixels either way, so it still has to work out which element you meant. A screenshot at 2am and a video at 2am both need someone to say this button, not that one.
The transport eats you too. I measured a 3066 x 28800 capture arriving at the model as 276 x 2600, so the text was under 3px by the time it got read. Video frames get resized on the same rules.
I went the other way with SlimSnap: mark the element, add one line of intent, and the agent gets JSON with the text and coordinates instead of a picture it has to squint at. Free, Mac, open schema. Different bet than yours, same enemy.