Technical Context
I love those small details that genuinely change the entire rhythm of work. In the Codex app, the annotate feature in the built-in browser turned out to be exactly that: instead of writing a wall of text, I just click on the problem area on the screen and run the next iteration.
If you look at this as an AI implementation for frontend, the win isn't in the model's magic. The win is that the loop of "see, mark, fix, check" has become shorter and simply faster.
I tested this scenario on typical UI tweaks: margins, button sizes, block positioning, and visual alignment with a mockup. Instead of a vague "make it neater," I provide a local signal directly on the interface. For the model, this is better than trying to extract a verbal description from me of what's already visible in the browser.
And this is where /fast mode comes in handy. In frontend, you rarely need one massive change; it's usually a series of small visual adjustments. When the response is faster, annotate stops being a cool feature and becomes a standard work tool.
Essentially, the built-in browser becomes a control surface for the agent. I look at the result, mark a specific defect, give the command to rework it, and immediately check the next pass. For slicing and dicing the frontend, this is genuinely more convenient than a purely text-based dialogue.
What This Changes for Business and Automation
Teams with many small UI iterations benefit the most here: landing pages, user dashboards, internal panels, and MVPs. Where a designer, PM, and developer used to send screenshots back and forth, a part of that cycle can now be compressed.
Ironically, those who expect the agent to deliver a "perfect interface right away" lose out. This combination is effective as AI-powered automation for quick refinements, not as a replacement for a proper design system and a clear component architecture.
I'd add an important caveat: this mode is especially useful when you already have a project's structure and clear UI rules in place. At Nahornyi AI Lab, we solve these kinds of practical problems: determining where AI integration actually speeds up a team versus where it just creates extra noise.
If your frontend is drowning in endless minor tweaks, you can set up a process where an agent takes over this routine. Contact us, and we'll look at your pipeline at Nahornyi AI Lab and figure out how to carefully build AI automation without just adding another toy for its own sake.