The Website That Builds Itself: Inside the Delivery Pipeline
Buy a website and, thirty seconds later, an AI wizard starts building it for you. Here's what happens in that thirty seconds — and why we engineered it the boring, reproducible way.
TL;DR
- When someone buys Prompt Your Site, the purchase mints a license, pins the buyer to a dated template tag, and streams that exact tagged zipball from GitHub. An agentic onboarding wizard then interviews the owner and finishes the site. Because the delivery is version-pinned, every buyer's build is reproducible — we know precisely what code they received.
- The pipeline is two decoupled apps — a marketing site and a buyer portal on separate databases — that coordinate through Stripe webhooks, each side idempotent.
- Version-pinning is the unglamorous decision that makes per-buyer support possible: a frozen snapshot beats "whatever was on main that day" every time.
Why does unpinned delivery cost more than it looks?
Ship every buyer the latest branch and you've signed up for a support problem you can't see yet. The main branch moves. Two buyers who purchase a week apart get different code, different bugs, and different behavior — and when one writes in for help, you're guessing which version they actually have.
That guesswork compounds. Every unpinned build is a slightly different product, and a studio operating real platforms can't afford a catalogue of one-off snowflakes. The cost isn't visible on launch day; it's visible three months later, in every support thread that starts with "it works on mine."
How a purchase becomes a running site
The pipeline turns a single checkout into a branded, deployed, owned website. It runs in four stages, and each one is designed so a failure can be retried without making a mess.
Purchase and license mint
Checkout completes, and a Stripe webhook fires into the buyer portal. The portal records a purchase, normalizes the buyer's email so a later sign-in links to the right account, and mints a license_id. The portal and the marketing site run on separate Neon Postgres databases with no shared code — so a hiccup on one side never corrupts the other. Both webhook handlers are idempotent, so a replayed event updates the same row instead of duplicating it.
Version-pin to a dated tag
This is the load-bearing step. Rather than pointing the buyer at the repository's moving main branch, the pipeline pins their license to a specific dated template tag. That tag is a frozen snapshot of the whole template at a known moment. From here on, "this buyer's build" has an exact, inspectable meaning.
Stream the tagged zipball
When the buyer claims their site, the pipeline streams that exact tagged zipball straight from GitHub — the frozen snapshot, not the live branch. Two buyers who purchased weeks apart each get the build their license pinned, deterministically. The onboarding checklist the portal shows them is generated from the web repo's own manifest, so the instructions always match the code they received.
Agentic onboarding finishes the job
The buyer opens their new repo and runs /wizard-setup. An orchestrator agent conducts a plain-language interview — brand, customer, design, products, blog — then walks the technical checkpoints and deploys. The owner never edits code; they answer questions, and roughly thirty Claude skills plus five brand-trained agents do the building. Those agents ship inside the repo, so the buyer keeps them.
Why version-pinning is the whole trick
Reproducibility is the difference between a product and a pile of similar files. Because each build traces to a dated tag, we can answer "what exactly did this buyer get?" without archaeology — which makes support, upgrades, and honest debugging tractable.
We prove the pipeline on ourselves. The site you're reading was built with Prompt Your Site, alongside analyticalants.com and hishighestfitness.com. Dogfooding isn't a marketing line here; it's how we find the rough edges before a buyer does.
What people get wrong about "AI builds your site"
The common assumption is that the AI is a magic box that spits out a finished site from a prompt. It isn't. The intelligence is in the scaffolding — a version-pinned template, idempotent webhooks, a manifest-driven checklist, and an orchestrated interview — that gives the agent a reproducible starting point and checkable steps. Remove that engineering and you get exactly the unmaintainable AI slop a technical buyer is right to fear.
Frequently Asked Questions
What does version-pinning a build actually mean?
When someone buys Prompt Your Site, the purchase records a dated template tag and streams that exact tagged zipball from GitHub. Every buyer gets a known, frozen snapshot rather than whatever happens to be on the main branch that day — so their build is reproducible.
Why not just give every buyer the latest code?
The latest branch moves. If two buyers purchase a week apart, unpinned delivery hands them different code and different bugs, and support becomes guesswork. Pinning to a tag means we know precisely what any given buyer received and can reason about it later.
Does the buyer own the code they receive?
Yes. Prompt Your Site is a one-time purchase, not a subscription. The buyer receives the code, deploys it on infrastructure they control, and owns it outright — there's no platform rent and no lock-in.
See the machine for yourself
We'd rather show the pipeline than pitch it. If you want a platform built this same disciplined way — reproducible, owned, no rent — start a conversation and tell us what you're trying to ship.
About the author — William Rodriguez is the founder of Analytical Solutions, where he builds and operates the platforms the studio sells the method for. He architected the Prompt Your Site delivery pipeline described above. Learn more about us.