A children's tabletop RPG existed as a plain text file. Its designer was too anxious to publish it. One AI session produced a 16-page print-ready PDF gift prototype, a live marketing website at littledonjons.com, and a working spam-protected mailing list.
Little Donjons is an original tabletop RPG for children aged 6 and up, designed by Dan Anderson as a derivation of the RPG Donjon. In it, one player acts as Dungeon Master and describes a world; the others play heroes — a dinosaur warrior, a good witch, a fairy princess, a robot adventurer, a pirate captain — and roll dice to find out what happens when things get tricky.
The game's core insight is that losing a roll is not a failure state — it's a storytelling opportunity. The losing side gets to narrate what went wrong. Consequences are gentle. No one is eliminated. Characters can only be written out if their own player agrees.
The game existed as a complete, playable .odt text file. The designer had no idea how to turn it into a product, and was too anxious about the publishing process to start. The goal of the AI session was to do that translation — and produce the first prototype as a personal gift, so the designer could hold the game in their hands.
The website is a single HTML file (~1.1 MB including both cover images embedded as base64). It runs on Hostinger static hosting with no server, no database, and no monthly cost. DNS propagated within two hours of upload.
The mailing list pipeline runs through three free services wired in sequence: reCAPTCHA v3 scores the submission invisibly → Formspree validates the token server-side and emails the game's creator → Google Apps Script appends a timestamped row (Date | Name | Email) to a private Google Sheet. If the Sheet call fails, it fails silently — it never blocks the user-facing flow. On success, the form hides and a download button appears pointing directly to the hosted PDF.
The PDF was built entirely from Dan's source files — no design software, no templates.
The AI parsed four uploaded files (.odt draft, two .docx reference docs) and used
Python's reportlab
library to generate every page, then
pypdf
to merge cover images and content into a single output file.
Palette: parchment background, deep forest green, gold accent. All colors, typography, and layout programmatically defined — no Photoshop, no Illustrator, no InDesign. The gift letter (page 2) was written by the AI based on the emotional context the human provided in the original brief.
The workflow here — parse source files, generate structured output, integrate services, deploy — is domain-agnostic. Twelve more games are planned using this exact method. The same loop applies to physical product development.
Feed a requirements .docx and a target format. The AI generates a structured datasheet using the same reportlab pipeline — with your branding, units, and section hierarchy. First draft in minutes, not hours of layout work.
This site was built from a text file and two images. A hardware product page needs a hero image, a specs table, and a signup form — same three components. Reserve the domain, supply renders and copy, capture early interest before V1 ships.
The Formspree + reCAPTCHA + Google Sheets stack costs $0/month for a typical hardware launch list. Same architecture, different fields — add a product variant selector, a region dropdown, a "notify me at launch" checkbox. Production-grade lead capture without a backend developer.
Engineers write in bullet points and jargon. Packaging needs plain language. The same parsing approach used on Little Donjons works on engineering documentation — AI extracts salient facts and rewrites them for the target reader. Cuts the copywriter brief to a conversation.
A working, professional-quality prototype in a single session. Code and HTML you can read, modify, and hand off. Architecture that separates concerns so pieces can be swapped as requirements change. A partner that handles boilerplate, remembers what you uploaded, and connects decisions made three exchanges ago.
Source material and domain knowledge. You must know what "correct" looks like in your field. You must push back precisely when something is wrong. You must be willing to go register three accounts and come back with the keys.
Library selection and service integration. Format translation (text → PDF → website). Boilerplate: forms, footers, field validation, credential wiring. The fire-and-forget Sheet logging pattern took one conversation exchange. The PDF took one session.
Anything requiring authentication, physical access, or real-world judgment. The AI cannot register your domain, evaluate whether your product is ready to launch, or know what the artwork looks like before you create it. You are the creative director. It is the production studio.
Highlighted = generated and executed by AI in-session. Plain = services the human registered. Zero paid software licenses. Zero backend server.
Little Donjons went from "a .odt file its author was too anxious to share" to "a live website at littledonjons.com with a download-gated mailing list" in a single two-hour session.
The designer's original work — the rules, the characters, the adventure — was unchanged and fully credited. What the AI provided was the translation layer: text file → formatted PDF → hosted product → working lead-capture pipeline.
This method is being applied to twelve more games using the same workflow. The pattern is repeatable. The cost is time and clear communication — not design software, not backend developers, not a production budget.
If you have source material and domain knowledge, ChessTrees can run this workflow with you. The tools are the same. The session length is the same. The deliverable is a working product, not a plan.