My 100% human-led, not-slop, AI-built new website.
I've spent most of my career building digital interfaces. For a decade, I led the TNEW build and scale-up (the ecommerce platform for the Tessitura community). I was part of literally hundreds of custom websites implemented on Tessitura data over the years.
So when I rebranded my AI consulting practice this month, I had a chance I relished: design my own thing, to my own standard, putting all my AI skills into practice.
My old site was a Squarespace prebuilt template (heavily customized but still a template). It fit what I had time and money for in 2025 at the start of this work. There is absolutely nothing wrong with prebuilt templates. They work.
But my full time job is AI practitioner and advising. I'd been using Claude Code a lot since it came out, and I wanted to know something specific:
How far can a person with strong design sense, tons of opinions, but no Figma experience and no formal design training get if AI is her collaborator?
This page is the answer. So is the site it's sitting in.
May 2026: the logo came first
The rebrand started with a word mark, not a website. I'd launched under my own name (kristindarrow.com) and that was great. But I knew I wanted a real brand for my cultural sector work with AI. I'd bought aiforartsandculture.com (and a bunch of derivatives) in 2025.
I am a big advocate of (literally) sketching your mind's eye on plain paper before you touch any form of technology. Typed, handwritten, drawn, spoken into voice memo... whatever. That goes for any creative project. I had the same rule back when I lived in Illustrator and Photoshop in my early career: create the idea straight from your brain first, then pick up the tool. Otherwise the tech will sway you silently, like a herding dog near a sheep.
So I sketched the logo on paper. Then I opened the AI in front of me, which happened to be ChatGPT 5.4 back in May, and gave it explicit instructions: the entity name, the positioning of the brand, the visual gist...super clean typography, strong word mark, no logo icon. Sans serif. It needed to scale from a conference badge to a slide header without falling apart, and it needed vertical and horizontal lockups in black-and-white and full color.
What I did not give it: the font. That's what I wanted help with, along with layout tension and proportion. (I've lost years of my life on 'fonting' (the endless pursuit of the perfect font.))
Could I have done this wordmark thing in Illustrator myself? Yes. I wanted to see what AI could do. Within minutes, I had something I liked and it saved me a lot of indecision and time.
After a few rounds I pulled it into Illustrator and built the real asset library: AI, EPS, SVG, every lockup, every treatment. Files I own, in formats a printer or a partner can actually use.
Then I looked at the word mark and thought: okay, so how does this work across my whole brand?
The design system in-formation
I heard Indra Nooyi (ex-CEO of PepsiCo) https://www.indranooyi.com/ say once that it's your "back pocket skills" that differentiate you in a career. So true.
One of mine is design sense.
I cannot draw or paint. But I can arrange and shape skillfully, and I know how to get close to what my mind sees. That skill has served me my whole career: PowerPoint templates for tech keynotes, board presentations, product demos.
Now, with AI, those design sense skills blend with decades of technology knowledge more powerfully than at any point in my career.
I would never call myself a designer (or an engineer). But I've worked literally in between the two disciplines for 20 years. That's a good definition of what product management is... the art of deciding what to build, what to design, and why. And then following up to see if it succeeded.
With this website, I was designing for myself this time (it always helps to have the client close by!).
So I knew what I was going for. Whether AI would meet me on that ground? Unknown. Sometimes AI astounds me. Other times it's remarkably frustrating, stubbornly unable to do something it did beautifully yesterday. The jagged edge :)
"Get your reps in" is still the best advice I have for gaining AI expertise. As I type this, I'm on minute 32 of Codex 5.6 Sol (Extra High) churning out a CMS build for this site. It's a dance between us, 50/50, and I find it both alarming how much time it takes to build well with AI and astounding that this technology exists at all.
The palette, the type pairing, the grid logic, the information hierarchy, the editorial stance, the positioning, the words, the impact: mine. Every one of those in-depth navigational roads... mine. Coming straight out of my brain, with LOTS of steering to keep Claude Code and Codex off less savory tracks.
What AI did was let me test directions and spike on ideas at a speed I still find astounding. Fifteen layout variants in an afternoon. Contrast checks on every color pair before I fell in love with one that couldn't meet WCAG AA standards. Layouts I could throw away without feeling like I'd wasted someone's billable week.
The real constraint on design quality is twofold: 1) your skill level, and 2) how many bad versions you can afford to make!
The design I threw away
If you've read this far, thank you, you're in for the whole story. I'll step it along.
The first real site design came out of ChatGPT right after the word mark was finalized. I put the logo in place, iterated on comps, landed on one I genuinely liked, then threw the static JPG comp into Lovable and asked it to make a real site.
I was astounded at how good it looked! It's still up at ai4ac.lovable.app (pardon the placeholder text it filled in without me asking). Last time I'd used Bolt or v0 or Lovable, they were novel prototype builders. Fun but not really that useful for real work. This built a real site I could actually imagine backing with a CMS.
I found my laurels and rested on them for a few weeks while I did other things, confident that I had a good design direction and it would all come together as easily as the logo and v1 build of the new design template had.
Then I started pouring real content into the v1 site. And the design system entirely broke. The whole thing lost its structure. It didn't scale. The more content I added, the more it looked like the markdown file I was pasting from.
I went at it with Claude Code, with Fable, manually in Lovable, trying to make the system hold together. As with all patching a big thematic idea with AI, it got worse and worse. Aimlessly adding H2 tags and horizontal rules on everything.
It was here, I admit to panicking a bit.
The site I'd planned to populate in time for TLCC in August was kaput. Zip. DOA. I had no option. So.
Six days before launch, I scrapped it and redesigned the entire thing.
That's either reckless or exactly the right call. It was right. A design that only works with fake content is not a design. It's a mood board with a URL. Shame on me for being overly optimistic; I knew it was unlikely to scale once real text and widgets and modules arrived.
Let me rewind to an almost identical juncture and what I learned then.
Years ago I led a rebrand and site build for Tessitura. I ended up abandoning an agency relationship after burning six figures in fees before moving to an agency that worked well. Why? Sheer frustration at the mood-board-meets-real-design-system stage. Exactly where I was now with my own site. The fancy agency was all flash and no splash. The design system that resulted from the new brand and logo was weak and wouldn't hold up under corporate use.
My financial stakes were much lower this time but the outcome was just as important to me.
Which brings me to the actual headline. Not "AI built my website."
AI made it cheap enough to throw the work away and start over when the work turned out to be wrong.
Same instinct, same call, six figures apart.
This time? Abandon what wasn't working and get on with finding something that does (otherwise known as iteration).
The rebuild is what you're looking at now, and it feels far more like me than the thing I abandoned ever could have.
This time? I built a real design system along the way.
What actually built it
So I run $100 plans on both ChatGPT and Claude and I bounce between Claude Code and Codex constantly.
On design for the project and the logo, ChatGPT won. Specifically 5.5, which is not their new hotness model (GPT 5.5 spanked 5.6 in my opinion). I spent a full day trying to get good design out of Claude, Opus and Fable both. Claude Design-- laughable. Nope. But by iterating with Codex in GPT 5.5, I was able to build out a robust design system over three days.
On code, I stayed in GPT/Codex. 5.6 Sol Extra High for the critical infrastructure and security work, then I could ratchet down the token burn on the more rote coding later in the build.
In building the site, I ran two Codex sessions. In the first, I built up a ton of back-and-forth context on exactly what I wanted from the CMS (content management system): what the CMS should and should not do on content atomization, the site design, the design of the widgets and modules and information graphics...the AEO (answer engine optimization) and GEO (generative engine optimization) goals, the security work. That session became my code architecture collaborator. I used its context window sparingly and well. Then a separate session(s) built the code in pieces using Goal mode, one goal at a time, spoon fed, with me checking each increment along the way.
Architecture in one session. Code writing in another. Me toggling between them.
A site built for humans and bots
First the humans. Things that are human-readable table stakes in 2026, and still routinely skipped:
- WCAG AA compliance, verified, not assumed. Contrast, keyboard access, semantic HTML, real heading structure.
- Every device breakpoint honored. Mobile, tablet, desktop. Not "responsive-ish."
- Information hierarchy that survives a five-second scan. If you can't tell what I do in five seconds, the design failed. (And please email me that fact. I'd be grateful, because I'm swimming in words and not remotely objective.)
Color was the genuine fight. The palette runs hot (black, white, yellow, electric bluebird blue, brand red) and some of the most beautiful pairs simply don't pass accessibility contrast ratios for body text. So the "hot red" and "hot bluebird"moved to accents, chips, labels, emphasis. Text-bearing fields got accessibility-compliant variants.
I hope nobody notices these decisions. That would mean it's working.
Built for agents
I spent a long stretch on something most sites skip entirely: structuring the content and data so machines can read it well. Something every one of your websites needs to focus on today.
A clean schema. Semantic landmarks. JSON-LD as a linked graph. Answer-first sections, because that's how content gets lifted into an AI answer. Content modeled in Sanity CMS as relationships, not just a pile of pages.
I don't need a public API for a site this slim. But I'm building the data structures now for the practice I'll have in three years, because retrofitting structure onto unstructured content is not a lot of fun. We all have plenty of that data catch-up to do. I didn't want to add to my own pile.
Ok, let's hit the uncomfortable questions
"Didn't you displace a designer and an engineer by using AI?"
No. And I want to be precise about why.
Before AI, I simply would have stayed on my ratty templated Squarespace site and thought the world was wonderful.
I would decidedly not have commissioned a custom site. That's the real counterfactual, and it's the same one facing most cultural organizations right now.
Resist AI on principle? Or figure out ways AI can bring newfound capabilities into your job, and your team.
I strongly advocate for the latter. I built my whole practice around that premise.
AI lets small and mighty teams (and solopreneurs) attempt things we wouldn't have attempted before, and do them well. When humans stay in charge of the outcomes and the direction. (And have dogged determination in abundance.)
"But these tools were trained on designers' and engineers' work."
Yes. Every layout pattern the model suggested rests on decades of work by hardworking designers, typographers, front-end developers and backend engineers who solved these problems first. It's 100000% true. And I'm not going to argue that the economics and ethics of that are settled. They aren't. Not by a long shot.
What I will say: the outcome is a site that is completely native to me. Hand-decided at every level, hand-sketched at many of them, with no template's fingerprints on it. It is something that would not have existed before for me personally. It feels like my own in a way I've never experienced with a SaaS product. (And I built a SaaS product!)
"Why not just use an agency?"
I wouldn't have. Cost being the biggest factor. But, I wanted to own the whole stack: my AI work keeps expanding my toolbox as a builder and a leader.
The finding-out was the point. Some of what I learned is on this page. More of it shows up in how I work with clients, which is the actual return on the experiment.
What this means for you
If you work at a cultural organization, the lesson is not "fire your vendor and build it yourself." (Though "build it yourself" is never bad advice for learning anything.)
The finding is that the limits have moved for each one of us. Work that used to require a budget you don't have is now reachable by a person with disparate skills and the patience to learn and iterate.
Production is no longer the scarcity. It's knowing what to build. Why to build it. Who to build it for. What good and done look like. And refusing to quit until it's both good and done.**
That's true of a website. It's also true of nearly everything else AI is about to touch in your organization.
Those are the skills I teach organizations to tap:
- Product thinking. What should we build? Why? For whom? Will it solve the problem we set out to solve? How will we know?
- Cross-functional work in a new way. Helping teams become innovation teams, lean into unknowns, and combine talents to push past prior limits.
- AI capability: How to develop competence and capability in the bleeding edge technology that is emerging and changing our world so quickly.
- Evolution of work: AI is shifting how we work, our teams, our lives. Leadership teams need to know how to support the right kinds of boundary-pushing, how to curtail the wrong kinds. And how to lead from exploration themselves. The skills that keep you pace with AI are the same ones that navigate any change widescale change.
I work with exec teams on the organizational side of AI transformation.
I work with frontline teams and individuals on AI enablement and technology integration into real life work.
If you want to talk about how AI can work FOR you or your organization, that's what I do.
I consider this website a deeply creative output of my own design process, one I've been running for decades. What changed is not the judgment. It's that I can now iterate sophisticated code and design at a speed I have never had before.
The choices are mine. The mistakes are mine. The sleepless 6 nights and 1000 saves in the CMS are mine. The speed is new. The expanded ability to custom-code a site of this quality entirely out of my mind’s eye? New.
Models Used
Logo and design direction: ChatGPT (GPT-5.4, GPT-5.5) Design system: Codex with GPT-5.5 Sol Code and CMS build: Codex with GPT-5.6 Sol (Extra High or High) Content and critique: ChatGPT (GPT-5.6 Sol), Claude (Opus 5) Tried, didn't make the cut: Lovable (first site design), Claude Code, Claude Fable (love them… Codex won on this project) Stack: Sanity CMS, Supabase, Vercel Vector production: Adobe Illustrator (human hands)
Last updated: July 30,2026