How toAgents & ToolsLenny's Newsletter · 2h ago
Improving interface design quality with automated subagent critic loops
Coding assistants often accept their own design outputs without questioning visual quality. Setting up an independent critic subagent to evaluate screenshots without seeing the source code creates an objective feedback loop that pushes design quality higher.
Try it yourself
- 1Set up a separate subagent using an advanced multimodal model to act as a design reviewer.
- 2Instruct the main coding agent to capture screenshots of the generated interface after each modification.
- 3Pass the screenshot to the reviewer agent without providing any implementation code or previous context.
- 4Require the reviewer to score the visual layout against high end design standards and suggest concrete improvements.
- 5Continue iterating automatically until the independent critic assigns a top visual quality score.
Copy this prompt
Iterate on this interface design using an independent multimodal model as a visual reviewer. Capture a screenshot of the current render, feed only the image into the reviewer model, and request a high level assessment comparing it to top agency work. Have the critic point out generic AI patterns and score the work out of ten. Continue modifying the code until the reviewer scores the visual output at 9 or higher.