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?
Anthropic and OpenAI are investing a lot into this space and are now competing directly with companies like Cursor. Cursor's biggest moat at the moment is their tab completion model, which doesn't exist in the Anthropic's and OpenAI's current offerings and is leagues ahead of Github Copilot's.
Antigravity is a VSCode fork that adds both Google's own tab complete and an agent composer, similar to products like https://conductor.build/. Assuming that Google doesn't shoot themselves in the foot (which they seem to like doing), we'll see if wrappers like Cursor / Windsurf / Cognition can compete against the big labs. It's worth noting that the category seems to be blurring, since Cursor has trained not only their own tab complete model but also their own agent model.
I haven't had much luck doing two completely different tasks at the same time. At net I think it slows me down due to the context switching.