Essay · AI
Where I find AI most helpful vs. where AI still falls short.
After a year of using AI in my design work, here's a breakdown of where I find it most helpful, and where it still falls short.
Where I find AI most helpful
Rapid prototyping
To quickly test ideas before investing time in full design or development. Here is more about my experience with rapid prototyping.
Headstart with boilerplates
One of the most useful things AI does for me is helping me get started.
Claude is useful for quickly filling out presentations, design briefs, and specs once I have a template or direction ready.
A blank page is usually the hardest part. Once there's a rough structure, it becomes easier for me to edit, refine, and add context.
I also use AI for formatting-heavy work that I'd normally spend too much time on manually, especially in Word or Slides.
Claude Code is also good at generating detailed wireframe specs and edge cases once enough context is provided.
Automate repetitive work
AI is also useful for repetitive work like:
- outreach emails
- meeting summaries
- release notes
- naming ideas
- labels and UI copy variations
I also use it for quick back-and-forth content exploration:
- “Give me 10 shorter labels”
- “Make this clearer for admins”
- “Reduce technical jargon”
It's still not very good at nuanced product language or context-heavy UX writing. I still find it useful to review complex passages with content designers.
Subject matter expert
I regularly throw large documents into Claude and ask it to explain concepts back to me.
The most useful part is usually the back-and-forth:
- why did it reach a conclusion?
- what evidence supports it?
- what sources is it relying on?
- what am I missing?
I've found the best results come when I treat it more like a discussion partner than a summarization tool.
Enabling designers to ship code
AI has also made small QA fixes easier.
Instead of filing tickets for:
- padding changes
- label updates
- spacing fixes
- color tweaks
- tab additions
…I'll often use Claude Code to make the change directly and submit a PR. This frees engineers up to focus on larger problems.
Keeping up with my work
I also keep a running log of weekly accomplishments throughout the quarter.
At review time, Claude helps me:
- group themes
- rewrite bullets
- identify impact
- structure the review more clearly
It turns performance reviews into more of an editing exercise than a memory exercise.
Overall, I find I lean into AI as a thinking partner, helping me compare ideas and pointing out blind spots.
The biggest value for me isn't automation. It's reducing the effort required to get started.
Where AI still falls short for me
AI is useful, but there are still areas where I need to step in heavily.
Context-heavy product thinking
AI struggles with decisions that span:
- multiple workflows
- organizational constraints
- technical limitations
- long-term product strategy
- historical customer context
It can reason about isolated problems better than systems-level decisions.
Truly different idea generation
AI is good at recombining existing patterns. It's not very good at generating genuinely different product directions without heavy guidance.
Trustworthiness
I never ship AI-generated work without verification. Hallucinations still happen regularly, especially in research synthesis and technical analysis.
Tone and judgment
ChatGPT still struggles with:
- nuance
- restraint
- tone
- editorial judgment
- intentionality
A lot of AI writing sounds polished but generic.
Visual craft
Claude design and Nano Banana are still not the best at:
- visual design
- iconography
- branding
- visual taste and maintaining visual consistency
Overall thoughts
Design used to require a lot of upfront thinking because building software was expensive. We tried to get things “right” before investing too much effort.
Now building is fast.
Which is wonderful, but also creates a new problem: feature bloat. We can build faster than users can keep up. The ability to build anything can turn into building everything, leaving users to figure out what actually matters.
AI helps with speed and ideas. But thinking, empathy, direction, and judgment are still human work. We need to be intentional about what we build.