Frontend Is Basically a Solved Problem for AI

If you have a clear idea of what good design looks like, AI can now build it for you. The taste is still yours; the implementation is basically free.

I’m not a visual person. Designing front ends and putting together slides are torture for me. But a lot of people care about good UI, and for good reason: it makes presentations land better, and being able to sell your own work can dramatically increase its value. Here are a few things I tried recently.

A few months ago I tried to replicate the Value Line report below and failed. This time, using the latest Claude model with a prompt built around the actor-critic framework, the agent could self-critique its way to an HTML version that matches the original PDF. It even pulls the underlying data from the SEC and Yahoo Finance and validates the numbers as it goes. The quality genuinely surprised me. Reproducing a PDF-like report in HTML is something I’d consider very hard for a human to do by hand.

I also tried building a Claude design system around my company’s PPT template. As risk analysts, we usually hand off our findings as Excel files, which are hard for end users to digest because they lack visualizations, highlights, and clear pointers. With the design system in place, I can now turn those same findings into a polished deck I’d never have been able to build myself.

The cost of building front-end visualizations is basically zero now. What matters is knowing what good looks like; once you do, AI can implement it for you. The next hill to climb is the backend business logic. So if you have a clear idea of how something should look, try handing it to AI.