The Process: A Collaboration Between Amanda Dahlgren and Claude Code

Visual Literacy in the Age of Machine Vision

Claude Code and I built a web application together. From an idea inspired by Trevor Paglen, to a cute little Claude Artifact prototype, to a frustrating day fighting Squarespace, and finally, finally to something that is actually useable. This page is a summary of the whole process: the idea, the fumbling and learning, what shipped, what got thrown out, and how long it actually took us to get to a live link on amandadahlgren.com. Straight from "git"* and my own memory.

Days elapsed

Build sessions

153

Commits

~72h

Logged build time

The commit/hour counts only start Jul 23 — the first two days happened before I even had a repo, so those are just what I remember.

*A quick glossary — I'm not a coder either, so I had to learn these too

Git — the tool that records every change Claude and I make to the code, like an infinitely detailed, timestamped edit history I can always look back through.
Repo (repository) — the project's folder, once git is tracking it.
Commit — one saved checkpoint of changes, with a note about what changed. Every "N commits" number on this page is really "N checkpoints saved that day."
ONNX — a file format for AI models that lets them run right inside a browser, on your own device, with no server involved at all.
Fluid Engine — Squarespace's own drag-and-drop page-building system.
Serverless backend — code that runs on someone else's computer instead of the visitor's browser, so it can talk to other services, without me having to manage a whole server myself.
CSS specificity — the rulebook browsers use to decide which style wins when two conflicting style rules both try to apply to the same thing on the page.
SEO fields — the title and description a search engine like Google reads to know what a page is about.
How this actually worked — Claude wrote nearly all of the literal code. My job was direction: deciding what to build, catching what felt wrong, testing it myself, and approving what shipped. When this page says “I built” or “I fixed” something, that's shorthand for that whole loop — not a claim that I typed every character myself.

Where it started

A week with Trevor Paglen

This whole thing traces back to one week: an Anderson Ranch master class with Trevor Paglen. I'd read his book, How to See Like a Machine: Images After AI, in the runup to the workshop. His throughline stuck with me: the glut of images today that are solely made for other machines to interpret, sorted and classified and acted on long before a human eye ever gets to them, if it ever does. I couldn't stop thinking about that gap. So this tool is my attempt to make someone else feel it, not just listen to me explain it.

"A machine reads for classification. A human reads for meaning." — the line I open the app with.

The timeline

Twelve sessions, one dead end, two pauses, one launch

From Jul 23 on, this is straight from git with real timestamps. The first two days (dashed cards) happened before I know I had to ask Claude to start a repo, so those two are just what I remember. The bigger the dot, the bigger the day.

What got built

The actual walkthrough, stage by stage

Forget the calendar for a second. Here's what the actual tool does today. First, you walk through the human experience of a deep dive into a photograph. Then I make you watch a machine do it to the same photo, quickly and poorly.

That straight-through path is the main event — but over time it's grown a handful of other ways in:

No webcam? You can decline the camera and still do the whole human walkthrough — Deep Looking even times how long you sit with the photo; you just skip the attention heatmap.
Let the machine start the research. On the Verify step, an optional reverse image search reports back only if the photograph has genuinely been published and written about — the real pages that carry it, never look-alikes or shopping results.
Critique mode. Run the same human steps on someone else's photograph, then — instead of the machine sections — download a written critique, meant as a gift for another photographer who's never seen this tool.
Request a critique. The flip side: if it's your own photo, a button opens a pre-filled email asking another photographer to run that critique for you.
Made it yourself? Skip the human steps entirely and go straight to the machine's read, downloadable as a PNG.

Where it stands now

It didn't stop at launch

It went live on my own site on Aug 3, but then it kept growing. It's since picked up a way through with no webcam at all, a mode for critiquing someone else's photograph (or asking them to critique yours), a mode for skipping straight to the machine reading of your own photograph, and, newest, an integrated reverse image search that will help you start your research into a well-known photograph.

→ amandadahlgren.com/visual-literacy