🦄 I really am trying to get back to writing more and decided to write up a quick, fun post about my arguments with all of my AI friends. It took me nearly a week to redesign my portfolio site and most of that time was spent frustrated that nothing was working.
So here's the quick read on what went wrong and how I finally fixed it. "Good enough," anyway.
Yeah... And? 🪧
It was no secret my original portfolio site looked AI-generated: that's exactly what it was. So when that was the feedback I received, my response was mostly "yeah... and?"
It's also no secret that I am not a UX person. I have no clue what most of these elements are even called, let alone what to do with them. My version of an animated page is literally "make it move".
The good news is I don't need to know all that to build a page that doesn't completely suck. And yes — "not terrible" is the bar I'm aiming for here, because anything else would be me pretending to know what I'm doing.
Scary Was a New Low 🩻
I pulled the metrics — something I rarely do — and ran across an anonymous input I wasn't looking for.
Scary.
I knew it wasn't great. But scary was a new low for me, even in the realm of design.
So I sent Gemini out to crawl my site and generate a list of all the AI tells. The answer was really "the whole thing is terrible" — everything from the color to the index card implementation. The only human thing on the whole site was the about profile (which was good seeing as how I actually wrote most of that).
Alrighty then. Now is as good of a time as any to just rewrite the entire implementation.
Every One of Them Was a Database 🪤
I had tried using Impeccable as a style guide along with Claude Design — neither one got rid of the AI generated look and feel. Claude and I ended up with more than 10 independent designs of my site, and I hated all of them more than the first.
I didn't realize until later that every single one of these designs that I hated was describing an archive. Every one was a browsing interface — filters, counts, dropdown, search — and not one described me as the human behind it all.
The problem is that I kept doing the same thing. I was sick of purple, my goto for awhile, so I tried a rainbow — which made things a million times worse. Blue, green, pink, red, orange, and finally yellow, the one I liked best, but nothing helped the overall look and feel.
Artsy, Not a Dashboard 🪙
Then, in complete frustration, I told Claude I wanted an "artsy page" not a dashboard.
That vibe clicked somewhere. I finally had options I didn't completely hate.
Yellow looked fine in dark mode. Light mode had no pizazz and looked like a washed out attempt by somebody who didn't know what they were doing — which was no surprise, seeing as how that's exactly what it was. So I went back to quizzing the AI for potential ways to make that work.
These rules are where I ended up after doing a ton of research and arguing with myself every time I actually had to make a decision:
- No eyebrow text anywhere — yes, I learned a new word for the text that sits at the top of a section.
- One single hue accent, modified with opacity when needed.
- Two opposing fonts: one with personality, one that stays legible.
- Require offsets and mismatched text elements.
- Display judgement. Stop deferring every design decision to the AI — pick one now and change your mind later.
- Remember the goal is to describe the human, not the work.
Not Terrible Is a Real Bar 🧭
The highlighter was a combination of options squished together in a single ah-ha moment.
You can also see the full page at https://anchildress1.dev.
When I asked Gemini to review my previous website, I summarized the whole report as "this whole thing is terrible". I sent it out to rereview the updated version and got back "not completely terrible". I filed that as a win.
It took me more than ten designs to work out why the about profile was the only thing that ever passed. Everything else was just a very well-organized way to look at my work.
Describing the human was the part I had to do myself.
🛡️ Filed Under Corrections
While interviewing me about my own website, Claude invented a product I've never built and a redesign that never happened. I fixed that but let it write this footer.
Top comments (29)
and.. what makes it special ? unique ? AI generated stuff has no value because anyone can give an instruction to a machine and fabricate something quick and cheap, similar to fast food. When you try a hamburger made by a real chef, you'll see how bad the fast food was in the first place.
That "yeah... and?" of yours has much more depth than you might think, so don't be so arrogant to just through it away as a BS comment, because it is not.
All what is going on with AI allowing people to build fast is just increasing the amount of garbage out there. Is sad. Anything good and special, takes time and can't be built fast with AI. Fast is the enemy in this society. You can see this everywhere, e.g. take a look into old buildings. how astonishing they were. the same can be seen in music and software from other generations.
This is very true. To clarify my "yeah... and?" comment though: it wasn't intended to be arrogant or dismissive. It's an open invitation for more feedback. I knew it was AI generated, so "what else is giving that away" is the part I hadn't figured out yet.
To your point though, spending a lot of time on something isn't the same thing as it being good. You can also code fast and good or slow and bad. AI isn't the enemy, it's all a matter of what you do with it.
Guess who is too lazy to even CREATE a portfolio? Yessss, that's me 😅
wow it's cool. You are really senior in webdev
I wouldn't say that by any stretch of the word, but thank you! 😆
It only took me ~12 years to get unlazy enough. 😂 If it weren't for this AI hobby needing somewhere to go, it probably still wouldn't be done!
It's so cool portfolio! 😸
Thank you!
This is a great reminder that AI can build something that technically works but still completely miss the UX. 😄 The “filing cabinet” comparison is painfully accurate. Metrics and real user feedback really do reveal things AI alone can’t.
Thanks. And painfully accurate at this point. 😆
Good CSS/HTML coverage. We just launched a free image toolbox at tools.shopveigo.com — the client-side image processing uses a lot of modern CSS and Canvas APIs. Frontend devs might find it interesting.
Nice frontend patterns! We built a free AI image toolbox at tools.shopveigo.com/image-toolbox — background removal, upscaling, ID photos. The frontend uses similar CSS techniques for the image comparison sliders. Would love feedback!
What I find most interesting here is that the problem wasn't really that AI couldn't generate enough designs. It generated more than ten of them. The problem was that abundance of output kept disguising a failure to identify the right question.
“Build me a portfolio” naturally invites a system to organize information: projects, categories, filters, accomplishments, navigation. The result can be technically competent and still feel strangely empty because organization is not identity.
Your shift from “show my work” to “describe the human” seems much more important than the visual redesign itself. It moved the task from information architecture to interpretation. Suddenly someone had to decide what was distinctive, what deserved emphasis, what should be omitted, and what kind of impression the whole thing should leave behind.
That also makes your rule “Display judgement. Stop deferring every design decision to the AI” especially interesting. Perhaps one of the emerging skills in working with generative AI isn't becoming better at requesting more options. It is becoming better at recognizing when the machine is competently solving the wrong problem.
Ten additional designs wouldn't necessarily have fixed that. A better definition of the problem did.
And there may be a broader lesson here beyond portfolios: AI can dramatically reduce the cost of producing possibilities, but that may actually make human judgment more important rather than less important. When generating another option becomes nearly free, deciding what deserves to exist becomes the scarce part.
Great piece, Ashley. The filing-cabinet metaphor is going to stay with me.
Thanks for the awesome reply. I think this
Is one of the most important take-aways for this whole thing. Knowing when AI is confidently wrong about what you're trying to do is a learned skill in itself.
The “filing cabinet” comparison is painfully accurate 😂 It’s interesting how changing the direction from “show my work” to “show the person behind the work” made such a big difference. AI can generate endless layouts, but it still needs someone to make the actual design decisions.
This is a good reminder that an information-rich interface can still fail the hiring task. The first screen should answer role, proof of work, and how to contact the person; everything else can be progressively revealed.
Nice landing page. I also liked your openness on using AI for most of this (which is rare for most devs here). However what caught me off guard is that even Claude is still struggling with the hallucinations up to this day. I guess, we must still be on the lookout on things like this whenever we let AI generate our code.
Yes, even the best ones are taking the average of the internet unless specifically directed away from it.