Scrawl
A visual frame is such a simple idea that I'm surprised more agent tooling hasn't led with it. We've all been there, typing out "the button on the left side of the header is misaligned" when a single screenshot would say everything. Scrawl basically removes the middleman between human perception and machine input, and that's where I think the real value sits. The harder the problem is to describe in words, the more this approach wins.
I'd want to know how smart the agent actually gets about the frame. Does it just grab pixels, or does it understand context, like what layer of the app you're looking at or what workflow you're in? That's the difference between a nice demo and a genuinely useful tool. If it can reason about what it sees, this could fold into debugging, QA, even design reviews. If it's just a screenshot uploader with extra steps, it's a feature, not a company.
Still, the instinct here is right. We're moving toward agents that need to see what we see, and the ones that nail that will feel magical. I'm curious to watch where Scrawl points its frame next.