The setup a one-night UI can skip.

A lot of AI-built projects die between “it looks right on my machine” and “someone else can open it.” The usual cause is a setup list copied from a company that already has users.

The list people inherit

Ask a coding agent to deploy and it will often propose the stack it has seen most often in production tutorials. An account at a cloud provider. A virtual machine. SSH keys. A domain name. DNS records. A certificate. A process manager so the server restarts. A database, even when the page has nothing to store. Each item is reasonable for a service that must stay up for strangers. Stacked together on a Tuesday night, they are how a landing page turns into a weekend that produces no link.

The decision worth making first is smaller. Who has to open this, and how embarrassed are you if it is down on Sunday? A page you will send to three friends has a different ops budget than a checkout flow.

What the small page actually needs

If the project builds to static files, the browser is the runtime. You need a folder with index.html and the assets it references, served over HTTPS, at a name you can copy. The certificate can be the host’s problem. The machine can be the host’s problem. You do not need a shell on that machine to change the page. You replace the files.

That covers a large share of what people vibe-code in an evening:

• a landing page or a personal site
• a clickable prototype of a product that does not have accounts yet
• a one-screen tool whose “data” is still in the design
• a form that only needs to remember a few fields, which can live in a table rather than in a database you administer

HTTPS matters even at this size. Browsers mark plain HTTP as not secure, and anything you type into a form should not travel in the clear. You still do not have to run Certbot yourself. A host that terminates TLS for a subdomain is doing the job.

What you are skipping on purpose

Skipping is a choice, not a failure. Write down the items you are postponing so the next session does not “helpfully” add them back.

• A cloud account in your name. Useful when you need private networking, custom runtimes, or a bill you can attach to a company. Useless as a rite of passage for a static page.
• SSH and a long-lived server. The moment you have a server, you have updates, disks, and a process that dies at 2 a.m. Static files do not need that job.
• A custom domain. A subdomain on the host is enough to text someone. Buy the name when the project survives the week.
• A second environment. One URL is enough until you are afraid of breaking the live page. Until then, deploy forward.
• Accounts and roles. If the only secret is “people I sent the link to,” a simple gate is enough. User tables come later.

The point where the skip expires

Add the heavier piece when the page starts doing a job the file host cannot. People log in as themselves. You store files, payments, or anything you would be ashamed to leak. You need a private network, a region, or a contract. Traffic no longer fits a small monthly cap. Those are good reasons. “The tutorial started with a VPC” is not.

The skill is telling a one-night UI from a system. Most of what an agent offers to configure is the system. Most of what you meant to ship is the UI.