Claude Code now opens a real browser to verify UI changes end-to-end, catching bugs like missing buttons that isolated backend and frontend checks miss. Agents need hard enforcement, not just instructions, to actually run verification. The tool excels at generating production code, designing architecture, and creating realistic test data for complex scenarios.
Having this was one big game changer. One thing that really changed a lot for me was when I told it to everything verify in the browser themselves. They have like with MCP, the connection where they can open a browser and click around. They click around, they even take screenshots. And so verify that the work they've done is correct. And before that, often they said like, oh yeah, it works totally fine, everything is finished. I opened the browser and I see like, okay, the button is missing or it's not connected to the backend or something because they only verified like locally the backend and maybe locally the frontend but not the whole interaction. So having this was one big game changer. As well as like having like a development process that it commits like small pieces of work where they— it always runs a lot of automated tests. So a lot of it's basically best practices of software engineering, but you need to teach the agent or cloud code to follow those practices. Here, another learning for me, very important learning, is you cannot trust it if you put it into the system prompt. First, I said, please always verify in the browser your changes. It always, or half the time at least, it found an excuse why not to do so. It always went around my guidance or it forgot about it. So I, after sometimes I really enforce it, I put a hook like, okay, you need to do it. Like there's no escaping from it.
And you need to verify probably yourself visually by your human whether those changes have been made. Yeah. Yeah. So where do you think Cloak Code helped the most?
I mean, the first one would be writing the code. I think the obvious one, I like the design and Claude is really good in designing, but also if you have the right framework to follow the design. And third, it would be test data. So I know from my old job, like when building model features, like half the time at least of building those features went into creating test sets for like niche niche supply chain configurations or something so that you ensure that it always works. And it's a lot of work to design those test cases. And like Claude is really good at building test data.
