Skip to main content
← Back to case studies

Applied AI · Web Experiments

One step closer to instantinaity.

In Quest of the One-Click Website

Pixel adventurers approaching a jungle temple labeled one-click website

Models on deck

Codex + GPT-5.2 + VSCode Copilot

Build pace

Record-time ship

Impact

Proved that multi-model collaboration can ship a conistently branded website pretty darn fast.

Request the private appendix for architecture, eval harness, and governance details.

Hello World!

We team-built this website with AI, but otherwise treated it like we would have any website development project back-in-the-day. The goal was to see how quickly we could build a static site if we already had the main components, like branding and logos.

The brand was designed by our team using standard design principles and tools. So, we had complete control of the pipeline from start to finish.

The models basically did the rest. With the assets in place and a clear, well defined blueprint (i.e. website prototype) the AI was more or less able to go from there.

It's pretty boring... but it works.

How We Assembled It

This was a lot like one of those baking shows you see, where they have all the ingredients laid out and a clear recipe to follow.

  1. Produced the brand and core assets, including a detailed website prototype (with consistent file naming, typography, and paragraph/character styles).
  2. Wrote a prompt to guide Codex 5.1 in generating a first draft from those assets, using the prototype as a visual map. Copilot produced a usable draft in ~5 minutes.
  3. Iterated with the team to refine content and layout—essentially the same collaborative design process as usual.
  4. Tasks that are typically fiddly—like getting CSS to match the design and supporting dark mode—were comparatively effortless. Understanding Design and Development principles allowed us to leverage AI more effectively.

This foundation allows for rapid iteration and reliable automation.

How We Deployed It

We lobbed the repo at Vercel the minute the gradients stopped flickering, and typedRoutes promptly cited us for a rogue string sneaking through the gate.

Instead of guessing, we installed the Vercel MCP server, piped the failed build through Copilot, and let it highlighter-pass every CTA that needed a typed destination.

Three edits and a fresh `npm run build` later, the redeploy cleared and the dashboard looked more synthwave equalizer than warning beacon.

Good telemetry beats lucky deploys.

What We Learned

We’re chasing the one-click AI website sasquatch, like many. And this is definitely NOT it.

But otherwise it's still pretty cool! The process and the system behind it are a solid step which we can build consistently with and that already has AI baked in.

Uh... And now a "block stacking game"

Uh, you know this is just an interim website we needed super-quick....

But, we're an "Applied AI" team, so we wanted something #future somewhere. Codex 5.2 built this (and wrote a draft on this case study) in a about six minutes, using consistent styling and branding. (Those blocks are our gen-u-ine brand colors!)

Arcade Telemetry

Lines cleared: 0

Mobile taps

Controls

  • ← → Move
  • ↑ Rotate
  • ↓ Soft drop
  • Space: Hard drop

Quest for the One-Click Website

We’re chasing the one-click website myth by building the repeatable foundation that can actually make it real.

Interestingly, all sorts of

Need A Calm Next Step?

Where to Explore

If you want the short version, start with a Tiny Seed. If you want receipts, browse the case studies. If you want the thinking, read the Field Papers—where we document patterns for getting data and workflows into a shape AI can actually use.