Mesopotamian Pottery: From Ramble to Research Paper
The Premise
Here is the exact prompt that started this project:
“I’m kind of wondering about like you know the evolution the evolution of Mesopotamian you know pottery and I mean you know how to how to how did that improve overtime or did it”
That’s it. No brief, no audience, no scope, no format. Just a thought spoken out loud.
What came out the other end: a 30-page academic report with Chicago Author-Date citations, sources from the British Museum and the Oriental Institute, a peer-reviewed bibliography, a live interactive website on a custom domain, and an email-gated PDF delivery system.
The Real Experiment
The question behind the question: how much signal does an AI agent actually need?
The conventional answer is “a lot.” Write a detailed brief. Specify the audience. Give examples. The more you put in, the more you get out.
That’s true for autocomplete. It’s less true when you’re working with an agent that has real tools, real sources, and the ability to ask itself better questions.
The ramble had something inside it: a genuinely contested question in archaeological literature. Did ceramic quality actually improve over 7,000 years, or is “improvement” even the right frame? That was enough to start.
How It Works
Research first. Three parallel agents ran simultaneously: one on chronology and named wares, one on trade networks and social context, one hunting open-access academic papers and museum image licenses. Primary sources only. Everything cited links to a DOI or a verifiable institutional archive.
Structure from the material. The report’s six core questions weren’t set in advance. They emerged from what the sources actually covered. The seventh section - proposed new research directions - came from noticing what the literature hadn’t resolved yet.
The PDF problem. The original plan was pandoc + LaTeX. No LaTeX engine installed. Rather than stopping, the process pivoted to md-to-pdf with a custom print CSS stylesheet: A4 pages, Playfair Display headings, running page numbers, a typeset title page. Indistinguishable from a professionally laid-out academic document.
The website. The report is dense by design. The website translates it for a general reader: a scrolling single-page presentation with an interactive 13-node timeline, CSS animations, and CC0 images from the British Museum and the Metropolitan Museum of Art. No JavaScript libraries. No build step. One HTML file, one CSS file.
The delivery system. The PDF is gated behind an email capture form. A Cloudflare Pages Function validates the address, builds a multipart MIME email with a styled HTML body, and sends it through Cloudflare’s email infrastructure. One email, one link, no spam.
The Numbers
- Chronological span: 6,900 BCE to 539 BCE (6,361 years)
- Phases documented: 13, from Hassuna through Neo-Babylonian
- Named wares covered in detail: 7
- New research questions proposed: 7
- Report length: ~30 pages, A4, print-ready PDF
- Website sections: 11
- Sources: British Museum, Oriental Institute (U. Chicago), Cambridge Core, Journal of Near Eastern Studies, Journal of Archaeological Science: Reports, Metropolitan Museum of Art
- Images: CC0 or Public Domain throughout
- Infrastructure: Cloudflare Pages, Pages Functions, Email Workers, custom domain
What We Learned
Scope creep, properly channeled, is a feature. The original ask was a research report. The website came from asking “how would someone with no background engage with this?” The email gate came from asking “if someone wants to keep this, how do we make that frictionless?” Neither was requested. Both made the output more useful.
The other thing: the question was actually worth asking. Seven thousand years of human craft production, administrative organization, trade across the Arabian Gulf, and social stratification encoded in fired clay. Whether “improvement” is the right frame for evaluating it is genuinely interesting. The report makes an argument. The website makes it accessible. The PDF makes it portable.
The Throughline
This project is a demonstration, not a one-off. The same pattern - vague question, parallel research, structured output, polished delivery - applies to any domain where the underlying knowledge exists and the question is genuinely worth asking.
You don’t need a five-page brief. You need a direction and a collaborator that knows what to do with it.
Current Status
Live at pottery.throughlinetech.net. PDF available via the email form on the landing page.