I rarely use the AI output as the final design

I have been using Claude Code for months to explore multiple design directions directly in Figma through MCP.

I rarely use the generated output as the final design. I take the useful ideas, remove what does not work, refine the flows, and continue the design work inside Figma. So far, this workflow has been far more useful to me than Claude Design.

Recently, I also started experimenting with Codex.

Visually, its initial design output has been less impressive. But it is useful for exploring edge cases, challenging assumptions, and acting as another brain during the design process.

That is where I believe AI currently delivers the most value for designers.

Most new landing pages are now generated, at least partially, with AI. I keep seeing startups launching with very similar layouts, visual patterns, and interactions.

This makes design craft, product understanding, and experience even more important. AI can quickly generate a polished interface, but standing out still depends on the designer’s ability to create something intentional, relevant, and memorable.

AI can support research, help structure product requirements, identify missing states, generate alternatives, and speed up execution. But the designer still needs to direct the process, evaluate every decision, and determine whether the proposed solution actually solves the right problem.

AI is here to stay. Our job as designers is to understand where these tools are genuinely useful, where they fail, and how to use them without outsourcing our judgment.

We are moving into a space where the quality of your thinking will matter far more than how quickly you can push pixels.