Agentation
The visual feedback tool for AI agents
607 followers
The visual feedback tool for AI agents
607 followers
Agentation turns UI annotations into structured context that AI coding agents can understand and act on. Click any element, add a note, and paste the output into Claude Code, Codex, or any AI tool.
Hey, I'm running a multi-agent Claude Code setup myself - one agent does UX/UI specs, another builds it in Astro. The tricky part is always the handoff: design agent says one thing, implementation agent hears something else. This looks like it might actually fix that. Does it handle stuff like when an element gets moved or renamed between design and build?
The Claude Code / Codex positioning makes sense but those tools have very different context window strategies. Does the structured output need to be formatted differently depending on which agent you're piping it into, or is the output generic enough to drop anywhere?
Text logs for agent debugging are genuinely painful — you're doing archaeology. The click-to-annotate approach makes sense for browser agents, but does this work for headless or API-only agents too? Or does the visual layer require a DOM to hook into?
Most Ai Coding workflow break down at feedback stage because we describe Ui issues ambiguously. This can help precisely locate the intent inside a Real Ui workflow and improve accuracy . wondering how it handles component mapping across larger dynamic Frontend.
Honestly, the biggest problem with AI coding agents isn't the code it's that the feedback they receive is too vague to act on. Agentation fixes exactly that. The MCP sync where the agent can actually push back or ask questions is what I'd want to test. I am curious how it handles conflicting feedback from multiple annotators.
Visual feedback for AI agents is something I didn't know I needed until I read this. Right now when my agent does something unexpected on the frontend I have to manually figure out what selector it acted on. Live DOM visibility during an agent run would cut debugging time significantly.
Congrats on the launch!
how does it handle errors or failed tasks?