My 180 on AI | Cosmic Ocean

I’ve definitely done a 180 on GenAI (which I’ll just call AI from here on out). For much of 2023 and 2024, I was fully part of the moral panic, which, I will say, is not completely unreasonable. The ethical concerns are valid. The moral panic is not. Then, toward the end of 2024 and into 2025, I concluded that it was in my best interest to actually get real, hands-on experience with the technology.

I’ve always been curious about how AI tools can complement art-making. I had a lot of questions, but of course professional artists are currently leading the Butlerian Jihad, so I wasn’t going to find my answers in their social media feeds.

Things are shifting, though. Slowly. Artists like James Gurney (yep, that James Gurney) have begun experimenting with AI, exploring how it fits into their creative process. I think we’re going to see a lot more of that by the end of this year.

Turns out, this technology is really useful at… everything. And it keeps improving drastically when it comes to image generation, video generation, and other creative scenarios.

It’s hard to find safe spaces to discuss this, though. People, especially the anti-AI crowd, are just super weird about it, and quite frankly, too militant for my taste. Over the past year I’ve thought I’d found safe places to talk about my exploration of AI, only to discover they aren’t.

And that’s okay. Whatever. I get it.

I don’t fully disagree with the ethical concerns.
There are so many unknowns, and the existential dread is real.

But I’ve also found myself feeling a lot of optimism about this technology.

How I Actually Use AI

When it comes to art-making, it’s been great to brainstorm concepts with Claude, develop lore for characters, get new perspectives on ideas. I’ll start doodling some of those concepts, feed the doodles to Gemini to help me visualize them more clearly, or use Gemini to complement my reference images.

When it comes to app development, I use Gemini and Claude in a very similar way: I love to discuss my ideas and vision, work together with the AIs to create design docs, or have them help me set up my dev environment. (As if I’m ever going to write a single line of code.)

Building Gastlichu.com

I wish I had documented the process better, but here’s essentially how it went.

I started by sharing with Claude the concept I had for my website. Described my vision for the style, the mood. Spent a significant amount of time just discussing the color palette. I wanted something in between space/galactic and underwater/deep ocean. I wanted the website to breathe, with constellations of stars and plankton floating around. At some point I described it as “tacky Geocities page from the ’90s, but fancy and modern,” and it was cool how Claude picked up on that and was able to work with me to find the right look.

I remember asking Claude to help me come up with search terms to find reference images on Google. Once I’d found the color palettes that felt right, we moved on to creating prompts for Gemini to generate more accurate versions of my vision. Then I fed those generated images back to Claude to create the actual color palette for the website.

That was the very first document Claude generated, a style guide.

After that, we started working on the actual design document for the website. I should mention: I set up all my AIs to always question, push back, and challenge me. So there’s always a healthy amount of responses along the lines of, “Your idea is not bad, but consider this…” or, “That’s one way to do it, but I’d recommend this instead. What do you think?”

I found that collaboration aspect (to the degree we can call it collaboration) to be genuinely fun and fulfilling. I loved just the planning. I think I spent an entire afternoon creating the design document, on top of the style guide.

Once both documents were done, I told Claude: I pay for Gemini Pro and Claude Pro, and I want to use AI to help me build this website. I also want to approach this as if I were a professional developer, so I need you to help me set up a professional workflow and dev environment on my machine.

Claude generated a strategy and a professional workflow, including a recommended tech stack:

Framework: Astro
Styling: TailwindCSS
Interactivity: Canvas API / Three.js
Version Control: Git + GitHub
Content: Markdown files

Every recommendation came with reasoning. Sounded good to me, so I went with it.

I ended up deciding to work mostly with Claude Code (for the first time!), but before I could start, I asked Claude to walk me through setting up my dev environment. Spent about an hour or two on that. Opened Terminal for the first time in years. Installed Node.js. Re-learned how package managers work. Finally figured out how to configure GitHub to both commit changes and push updates to my live website. Set up Visual Studio Code, which for some reason I keep open despite not writing any code. :P

Once all of that was set up, we broke the development strategy into phases, documented everything into an onboarding document for Claude Code, and I was finally ready to go.

Was the whole thing overkill? Could I have just started working directly with Claude Code? Honestly, I don’t care. I’m happy I did it this way, and this is how I plan to approach all future projects.

I was very hands-on. Phase by phase, testing the website, identifying issues, flagging them to Claude Code, fixing them, until the site was finished. Took maybe four to six hours total.

gastlichu.com

The Result

I’m quite happy with it. It’s not perfect. It runs well now, but there were some optimization issues that were tough to fix, and some concepts that didn’t work as well as I’d hoped.

You’ll notice the site has two menus: a traditional burger menu and a “Constellation Menu.” The constellation was supposed to be the primary navigation. You’d open a star map and navigate to each star, with each one representing a section of the website. But after a friend played around with it and was completely confused, I made the burger menu more prominent.

If you do use the constellation menu, notice that when you navigate between pages the stars in the background briefly keep some of their momentum. The effect is too subtle, though.

Something I am proud of is the galleries. If you go to any of the galleries and click on a photo to view it full-screen, you’ll notice a zoom feature. I’ve always been a little annoyed at how gallery images just open at full size with no control over zoom level. So I added zoom controls. It took several iterations to get it working smoothly (it used to be quite buggy) but I think it’s pretty solid now.

What It Means to Me

The whole experience is what made me switch to optimism about this technology, at least on a personal level. I want to be specific about that qualifier: on a personal level. I’m not dismissing the concerns around training data consent, labor displacement, or the concentration of power in a few companies. Those are real, and I don’t think my positive experience building a website cancels them out. Both things are true at the same time.

I also want to address the “not writing a single line of code” thing, because I know how that sounds. What I actually do is define the vision, write the specs, architect the workflow, and then test and break everything the AI produces until it meets my standards. I’ve spent nearly ten years in QA learning how to systematically find what’s broken. Building with AI turns out to be that same skill applied in reverse. I’m not skipping the work. I’m doing different work.

I only wish there were more people openly sharing what they’re making and how they’re approaching the process. But the taboo is too big right now.

Here are some other things I’ve made in collaboration with Claude:

Séance Sphere: A virtual pet demo on web
An iOS port that runs natively on iPhones, built using Claude Code + Xcode
Build a Rocket: A rocket builder game designed by my nephew

There are also some games I made earlier with Gemini, but those are no longer on my website.

I don’t know if any of this will ever replace my day job. I don’t know if jobs as we know them will even exist in the future. But I’m tired of waiting for permission to be excited about something. So I’m just going to keep making things and see where it goes.