We built the site with AI. Now we need humans for the rest.

We built the site with AI. Now we need humans for the rest.
— by DFT

Behind every page on this site — the vetting application, the Foundations curriculum, the Yes/No/Maybe checklist, the incident reporting workflow, the mentor matching, the badges, the staff-role training — I want to be straight about something: an AI wrote most of the code, working alongside me as the architect. To be exact about the "we": it's one person and one collaborator, and the collaborator is Claude, an AI. It's credited on 78% of the commits behind this site. I made the decisions and set the direction; it did the bulk of the typing.

This was the only way to get here. A platform that does what this one does — encrypted vetting records, audit logging, role-based access, scheduled emails, automated quiz grading, an end-to-end-tested integration with our calendar and group tools — is the kind of thing a small engineering team might quote a year of work and low-to-mid six figures to build. I want to be careful with that comparison: it's an industry-rate estimate for the scope, not a receipt. What I can prove is the scope, the timeline, and who did the work.

To put real numbers on it — and these are measured from the actual codebase, not rounded up to sound impressive:

• About 90,000 lines of hand-written application code (roughly 148,000 if you count every physical line across PHP, JavaScript, CSS and configuration). None of that is borrowed or auto-generated boilerplate — it's the vetting engine, the LMS, the story engine, the encryption, the plumbing.
• Around 276,000 lines total once you add the database migrations, the quiz banks, and the Tales storyline. Most of that extra is data and content, not logic — but it's all part of what makes the site work.
• Over 400,000 words of documentation and curriculum. The member-facing handbooks, policies, FAQ and glossary alone run about 65,000 words; the Foundations and Skills lesson bodies add roughly another 345,000. That's not "novel-length" — it's several novels' worth.
• Every code change ran through automated checks — well over a thousand individual CI runs across the build so far, catching broken syntax and type errors before anything shipped.
• Behind those sits a 41-file end-to-end test suite, backed by dozens more helper scripts, that walks the whole platform: submitting applications, grading quizzes, playing through the Tales story, filing test incidents, and running the vetting flow start to finish. By hand, one full pass is the better part of a day — and it runs every time anything changes. That's hundreds of hours of manual testing the machine absorbed so I could move fast without breaking what already worked.
• All of it built in about seven and a half weeks — 52 days from first commit to now, worked on 42 of them.

I had a budget of near-zero, a small community, and a timeline measured in weeks. So I built it with AI as my coding partner. One person, with one AI collaborator, shipped the kind of platform that would normally take a team the better part of a year.

What that's gotten us

• A vetting workflow that actually screens applicants — built in days, not months.
• Encrypted vetter notes (AES-256-GCM, decryptable only on our own server, with click-to-reveal and a full audit trail) so vetters can document concerns without exposing anyone's private information.
• An incident response framework, with cross-venue safety alerts and reporter privacy baked in from the ground up.
• The Foundations consent and safety curriculum — lessons, quizzes, badges, and automatic group memberships the moment you pass. Printable PDF handouts too, so you can take any class home with you.
• A Yes/No/Maybe checklist and a Scene Negotiation tool to compare interests with a partner before a scene.
• Tales — a choose-your-own-path story that teaches consent and community norms by letting you live the decisions, not just read about them.
• Mentor matching — newcomers paired with an experienced member and connected privately from day one, so there's someone to ask the awkward questions while you wait.
• A member home base — log in and see exactly where you stand: your vetting status, the classes you have left, and what's coming up.
• A volunteer staff-role workflow — apply, train, take a quiz, get approved — so we can grow our team without losing track of who's been trained on what.
• A one-tap quick-exit button on every page and a full accessibility pass — keyboard navigation, screen-reader support, and a WCAG-checked contrast pass across a dozen visual themes you can switch between.
• And the small delights: celebration overlays when you hit a milestone, birthday touches, and an "Ask OTT" assistant that answers your questions and hands you off to a real human whenever it can't.

Now we need humans for the rest.

Here's the thing the AI can't do: be the community. It can't greet a nervous newcomer at their first munch, or notice when someone's quiet at the edge of the room, or sit with a person walking through the day after a hard scene. It can't teach a rope class, run a discussion, mentor someone through their first year, or make a judgment call about safety that needs a human heart behind it.

The software is the easy part now. What we need is people:

• Welcome crew — the friendly faces at munches and first events.
• Educators — anyone with a skill worth teaching, from the fundamentals to the specialized.
• Mentors — experienced members willing to be someone's first point of contact.
• Dungeon monitors and event volunteers — the folks who keep our spaces safe and running.
• Vetters and response-team members — the careful, discreet people who do the hard, unglamorous work of keeping this community trustworthy.

If any of that is you — even if you're not sure yet — reach out. Apply, or just message us and ask. The machine got us a running start. Everything that makes this an actual community, and not just a website, is up to us.