Building REPLICATE

Tom Bowyer

— 2 min read
Building REPLICATE

If you have spent any time with AI coding agents, you already know the obvious wins: scaffolding projects, writing glue code, churning out boilerplate.

I wanted to push on something different. So I built REPLICATE, a digital museum of the history of self-replicating code, spanning seven exhibits from 1949–1988.

I wanted to see if Claude Code could help build an interactive experience that feels intentional. Not “a demo page that works”, but something with mood, pacing, and little details that feel good.

What REPLICATE is

At a high level, it’s a timeline-driven museum experience:

  • A landing page that frames the story as “seven exhibits, four decades” and invites you to enter the timeline.
  • A “Digital Posters” collection, where each exhibit gets its own canvas-based generative art poster.
  • A “Wallpapers” collection with static renders of the same generative art in 2K and 4K.

How I worked with Claude Code

My approach was closer to directing than prompting.

Instead of “build me a site about X”, I treated it like:

  1. Lock the vibe first
    Typography, spacing, palette, and a few UI rules. Once that’s stable, everything gets easier. You stop yelling about aesthetics on every page.
  1. Make interactivity deterministic
    Generative art is fun until it’s chaotic. Seeded randomness and repeatable outputs matter, especially when you want posters, wallpapers, and thumbnails to feel like part of one system.
  2. Build in tight loops
    Small edits, refresh, adjust. The agent is fast, but your eyes are the test suite.
  3. Use “structure prompts” and teams when complexity spikes
    When a change started to sprawl, I’d stop and ask for a plan, and deploy a team to accomplish it in parallel.

The stats (pulled from the Claude Code sessions)

This project was built across four Claude Code sessions in about 8 hours, roughly 9:58am to 6:00pm (America/Chicago).

From the session logs:

  • 4,613 recorded log events across the day
  • 589 explicit tool calls
  • Tool calls broke down roughly like this:
    • 170 edits
    • 113 reads
    • 88 greps
    • 86 bash runs
  • Total tokens processed = 99,297,646

Curious what we're building? Sign up for a trial at https://dashboard.turen.io/.

Read more