Designing with AI.
Pick the tool for the job. Then actually use it. Reading a release is not the same as knowing the product.
I don’t think there is one “AI workflow” for designers.
The more useful question is simpler:
What job am I trying to get done?
Research, making, remembering, synthesizing, prototyping, documenting — those are different jobs. They do not all need the same tool, and forcing them into one chat window is usually where the work gets sloppy.
So I’ve started treating AI less like one assistant and more like a set of assignments.
There is not one AI. There are jobs.
I care less about having a favorite model than I do about knowing what role each tool is actually good at.
NotebookLM holds interview notes and findings so marketing and other teams can ask what we already learned instead of losing it in a folder.
v0 and Lovable help me get motion, hierarchy, and a concept in front of stakeholders quickly instead of building a theater in Figma first.
A Claude loop catches decisions after important meetings and returns a weekly digest I can hand up. The point is not the model. The point is a job the week can feel.
Industry literacy without use is just a feed.
New models show up every week. Tracking them matters.
Using them matters more.
Starting is usually the hard part. Once you are actually inside the product, you can tell what is genuinely new, what is dated, and what only sounded impressive in the launch post.
That is the difference between knowing that a product exists and understanding what it changes for the person using it.
Instant is a good example for me. After signup, it reaches you through iMessage or WhatsApp.
The question I ask might be the same one I would type into a browser-based chat. But the channel changes the relationship.
It feels more like contacting an assistant than visiting a webpage.
That is not a model observation. It is a design observation.
I asked for design to work in code. We started small on purpose.
I convinced our CTO to let design use Claude Code.
But we did not begin on the flagship product and pretend design had suddenly become a second engineering organization.
We started with internal tools and backlog items — places where the team could learn by doing without creating unnecessary risk.
Learn the workflow before asking the organization to trust it on the most important surface.
Working in code is not permission to invent new patterns because generation made them easy.
If the workflow creates more review burden than value, the experiment is not working yet.
The contract with engineering is part of the design.
Keep integrity. Do not create extra cleanup. Do not ship mess because we wanted to be in the repo.
I am deliberately not publishing the internal workflow here. The decision to change how design participates in the build is the part I want to preserve.
This portfolio is part of the same habit. I used AI in code on purpose, not as a demo of tools.
It is to make the work better enough that the team can feel the difference.