There have been discussions about whether engineers and designers should design directly by building the design in code using AI Agents, or keep iterating in Figma.

Both approaches have their pros and cons, and I am biased because of my software engineering background. Figma gives more freedom to explore and go beyond creative limitations at the expense of taking more time per iteration and splitting the building process across two different mediums (and teams in most cases).

Designing in code directly with a good AI agent can be tempting, since implementation constraints would always be considered, with less friction in getting the design to production. However, anyone who tried doing that knows that after a couple of iterations, you get stuck with the model bias, and each change or iteration brings you closer to frustration.

However, given how much better the models have gotten, especially with Opus 4.5, doing low-fidelity design iterations in ASCII style has bridged the gap for me. I'm now utilizing the iteration speed with the AI Agent, I'm close to the codebase, and I'm having more freedom to explore without the limitations of the code implementation.

This take is heavily biased by my engineering background. I understand that many designers have their workflows already set up to be faster. But for generalists like myself, I'm constantly optimizing for designing the best possible experience for the user, in the most efficient way, with the toolkit and workflows that utilize the most of my experience and avoid its gaps.