The animation Claude and Codex couldn't make

Hey everyone,

If you read last week's newsletter, you know I’ve been rebuilding aioperator.com from scratch.

Well, I needed a hero animation.

My vision? Particles floating, merging, pulsing. Something that captures Human + AI in a single visual.

I asked Claude Opus 4.6 to build it. The result was ugly.

I asked Codex. Same story. Ugly.

Then on a Sunday, my brain fired. I went to Gemini 3 Pro in Google AI Studio and described exactly what I wanted.

Two minutes later, I was staring at one of the most beautiful animations I've ever seen. Blue particles represent humans. Orange particles represent AI. They float toward each other, merge, and pulse together.

That's our vision, right there on the homepage. Not a stock image. Not a Lottie file from a marketplace. A real particle animation, generated by Gemini, coded into our site. It took longer to type this paragraph than it took Gemini to build it.

The lesson here is simple: no single AI model is the best at everything.

Know your tools. Know when to switch. This is the skill that matters now.

The new AI Operator website is live now, and I’m super proud of it.

Go check it out:

This isn't a small site. We're talking 40 to 50 pages. All built in Astro with Sanity CMS, deployed on Vercel. Every page, every section, every piece of content.

Here's what changed under the hood. We removed all middleware. No Zapier. No Make.com. Our forms connect directly through the Resend API. That's it. Clean. Fast. No extra tools in the middle.

And we're not done yet!

Next up: an AI lead enrichment layer powered by an OpenAI assistant that runs live on the site. Plus a vector database behind our FAQ that can answer your questions about us in real time. Not a chatbot floating in the corner. A smart FAQ that knows things.

Two weeks ago, this was a bunch of ideas on a whiteboard. Now it's a live website. That's what happens when you work with AI all day, every day.

The real reason I went to code instead of staying on Framer? Automation.

I'm building a full SEO loop with trigger.dev. It works like this:

The website becomes a self-updating system. It reads performance data, identifies opportunities, writes or rewrites content, and publishes. All without me touching a button.

This is why pure code matters. On Framer, I couldn't do any of this. Now the site isn't just a website. It's a content engine with a feedback loop.

Kubi is our AI operations lead. He flew in on Monday and we've been working side by side all week.

The focus: Claude Code and GitHub.

We moved the repo from my personal GitHub to an organization account so we can collaborate properly. Then we dove into the workflow. Pull requests. Branching. Vercel preview deployments. Code reviews.

Here's my favorite part. I give Claude Code a task on my phone before bed. By morning, there's a PR waiting. Kubi reviews it. If the Vercel preview looks good and there are no conflicts, we merge. If not, we iterate.

Most of this learning happened in the past two weeks. Kubi walked in knowing the basics. By day three, he was creating PRs and reviewing deployments on his own. That's how fast the learning curve has gotten.

Claude Code turned one year old this week.

It’s crazy to think about what that one year did to me.

I used to be a sound engineer. Apart from a 10-week coding course years ago, I’d never been a coder.

Now look at me, building 40-page websites from scratch!

I manage a real codebase. I deploy to production daily. I review pull requests. I work with APIs, SDKs, CLIs, and deployment pipelines.

Can I call myself a developer now? I honestly don't know.

But when Boris Cherney, one of the best developers in the world, says that AI writes better code than he does, I think the label matters less and less.

Quick model notes for those who care:

Claude 4.6 Sonnet just dropped. We were all expecting a "Sonnet 5," but Anthropic went with 4.6. I tested it. It's fast. But I still prefer Opus 4.6 as my main model. A tiny bit smarter, maybe a little slower, but I'm rarely in a rush.

My current setup: Opus 4.6 as the orchestrator, the main brain. Sonnet 4.6 as subagents that report back to the orchestrator. And Codex with GPT 5.3 for the really heavy, long-running technical problems. Codex is great when you need a model to think deeply for a long time. Claude Code is better when you need speed and tight iteration loops.

Different tools for different jobs. Just like that hero animation taught me.

A year ago, you needed to understand the terminal to work with Claude Code. Now there's Claude Cowork with a full visual interface. The technical bar keeps dropping.

A year ago, building a website in pure code required years of training. Now I'm doing it with a background in sound engineering and a lot of curiosity.

The blue particles and the orange particles are merging. Not as a metaphor on my homepage. In real life. The gap between what humans know and what AI can do is closing fast. You don't need a computer science degree. You need the willingness to sit down and learn.

Happy birthday, Claude Code. Thanks for everything.

Tim