ai can make your design process better
How do you use AI in your design process?
AI completely changed how I do things compared to just a year ago. Here I share some of the more basic things that worked for me, and hopefully, could give you some ideas and inspiration. Especially if you aren't sure where to start with AI in your design process, feel overwhelmed, or worry about falling behind. I also welcome you to reach out and share your use cases.
Automate
tl;dr Find the friction that sucks your time and doesn't inspire you. Automate it.
In the summer of 2022 I joined a startup that was building a white-label product (you sell your product to your clients who then customize it and publish it as their own) for retailers. One of the friction points was producing color themes for clients, the team had to manually create multiple color schemes for different brands to show how the product can look like if they partner with us.
I didn't particularly enjoy looking how talanted designers wasted days on something that could be automated. So I learned how to build Figma plugins, spending my free time on JavaScript courses and digging into the Figma API and open-source plugins. I build the plugin that solved the issue, saved the team months of work, and then became a self-service platform for clients.
AI now lets you build things without spending months studying JavaScript. The Figma agent can create plugins and handle tasks right inside the app. External agents can control Figma too, using it's MCP server or Figma Console MCP, depending on the task and plan.
Here are some handy things I did using agents in Figma: I previewed my mockups with different font sizes to see if the layout breaks, updated spacing and corner radii for a new grid on multiple screens at once, and mass updated font styles. I also used them to find leftovers from old libraries, and to build simple prototypes and animations.
If you are ready to explore more, try Paper. It's agentic-first AI tool. I found it incredibly useful for co-designing with AI. One of the use cases for me is I sometimes ask an agent from my phone to draft a bunch of ideas in it, and then, when I have time and at my computer, I just open Paper and review the work.
More on these tools:
Discover
tl;dr AI is good in research. Use connectors and MCP servers to connect to your analytics tools to find answers faster.
My research used to mean jumping between Google, Reddit, reviews, and analytics, often with an analyst's help to make sense of the data. All of that took time, even without talking with users.
When deep research tooling was introduced, it became my go-to for initial analysis. Now I can ask a model to research a topic, browse the internet, and gather information for me to review. Besides web research, it can connect to the product analytics tools I use, parse and make sense of the data there without me needing to ask an analyst for help every time. Computer use and browser extensions also let models interact with websites that don't have MCP servers yet.
Recently, I've also found myself using Grok Bot more often for research because of how well it analyzes X, which is a goldmine of opinions and ideas.
I love talking to users and wathcing how they use the things I build. Without human connection it's easy to lose sight of why the work matters. AI helps me prepare for these talks so I spend less time guessing and get more value out of every conversation.
Ideate
tl;dr Sketch ideas and prototype with AI so that you don't have to start from a blank page.
Sometimes I just stare into space, playing through the possible options in my head and searching for the right solution. Sometimes I like to sketch by hand. Sometimes I browse Mobbin or revisit Dribbble, hoping to find inspiration.
When I have a vague idea, I often ask AI to build a prototype with ChatGPT Sites, Claude Artifacts, or sketch an idea in Paper. Interacting with it gives me a better sense of what I want and what it could become.
Sometimes ideation needs more constraints or product context. For those cases, I've found it useful to use Claude Code's /design skill right in the repo I'm working on. It can use context from the codebase and Claude's design tools to generate ideas on a separate canvas.
Another workflow I like is asking an agent to recreate a screen from my codebase in Paper. I can tweak or completely redesign it there, then ask the agent to implement the result back in the codebase. This lets me play around on a canvas and test a few ideas without manually recreating everything.
Craft
tl;dr you can build and polish details of your project directly in your product's code, and never say again that engineers are to blame that the product don't look as good as you designed it.
When I know exactly what to build and want to focus on the details, I usually switch to code and work directly there. I'll open a repo in Cursor, Codex, or Claude Code, preview the app in a browser, and tweak things through conversation. Being able to edit code manually helps too, especially when I want to get a detail exactly right.
Another useful tool here is Josh Puckett's DialKit, which adds live controls to a project. An agent can wire up controls for animation, spacing, colors, and other details, making them easier to adjust. Josh shares more about his approach through Interface Craft and on X, so don't be a stranger und subscribe.
Skills
tl;dr Use others's AI skills to make your work better. Learn from them too.
Sometimes I'd buy a book, say, on healthy eating or burnout, and already feel good: I'd made some progress. Then I'd put it on a shelf and return to my life and old habits, with a wonderful feeling of accomplishment. With AI skills, that shortcut can actually help. Once I install a relevant skill, the model can apply its guidance to my work.
There are some fantastic design skills available out of the box. My personal favorites are the design engineering skill packs by Emil Kowalski and Jakub Krehel. The Figma agent also supports custom skills, but I haven't tested those myself yet.
Build personal tools
tl;dr You can create tools for AI models that they can use to produce more reliable results where it matters for you.
Skills are just the beginning, though. At some point, I realized I needed more control over the outcome and more confidence in the model's results. I find it hard to trust a probabilistic model to get things right when I need precision.
Color contrast, for example, needs to be measured. A model shouldn't guess whether a pair of colors meets a contrast threshold; it should be able to calculate and test it, but to do that, it needs tools.
The easiest way to provide a model with a tool is by creating a CLI-app, an app with a text interface that a model (and a human) can interact with through terminal. Pair it with a skill that explains how to use the app and, woo ah la, the model can do the math, or get deterministic answer on what component to use in which situation.
!!! Getting back to the color contrast, I started building tools that AI could use to do the math. My favorite is APCA CLI, which checks text contrast using APCA and takes font size and weight into account. !!!
Build your own design software
At some point, I found that existing software like Claude or Codex didn't quite fit my workflow. I wanted to keep the benefits of my subscriptions and the skills and tools they already supported, without building my own coding agent from scratch.
So I decided to build Trezi, the 'design mode' I wanted my coding tools to have: rulers, inline edits, and a way to slow down animations to review how they move. It works with my Claude and Codex subscriptions.
Evolve
For me, the important thing is to stay focused on what I want to make and keep learning as I go. I don't learn well from courses. I learn by doing, and the best way for me to learn something is to face a problem I want to solve, or find something that annoys me, and try to fix it.
Here are a few more resources to explore, including some I mentioned above.