Ask HN: How do you verify front-end code in agentic LLM coding loops?

Agentic coding loops work well when the LLM can run tests to verify its work. For backend code, this is straightforward - write a test, run it, iterate. For front-end work, I haven't found a good workflow. I've tried Playwright MCP and Google Antigravity which has a Chrome integration with mixed results. How are you closing the loop on front-end verification?