How to put a system built with AI into production

A growing share of the projects that reach us already have a repository, screens, and even a local database. They were written with help from programming agents, in a few days. The next question is always the same: how does this actually go live, with the company's domain, without exposing secrets in a file in the developer's folder, and without depending on a laptop left on.

Writing code and operating an application are different jobs

The AI agent speeds up what is in the editor: route, component, prompt, shallow test. It does not replace choosing a server region, backup policy, certificate renewal, environment variables, queues, logs, and what happens on Saturday when a model API changes its rate limit. Production is that set of details, repeated.

That is why a system that is “ready on GitHub” is still not a product. What is missing is the path Code → repository → environment → domain. At SISNAWEB that path can be: GitHub, panel, our server or the client's, deploy, application live.

What is usually missing in a project built only in the IDE

• A managed database, not a SQLite file in the project folder
• HTTPS and a domain, not just a local environment with a certificate warning
• Secrets outside the code (LLM keys, tokens, passwords)
• Backup with a tested restore
• A process to update versions without handmade SSH every Friday
• Minimum observability: is the application up or not?

AI applications ask for a little more

Beyond the site or the API, you add storage for documents, sometimes a queue for embeddings, model-provider cost limits, and clarity about which data can go to which API. Agents, RAG, and backends that call an LLM break in ways an institutional site does not: long timeouts, a blown account, a stale index.

Do you need to have your own server?

No. You can use SISNAWEB infrastructure or keep the account with the provider in your name — Hostinger, Contabo, Hetzner, or another compatible one — while technical management stays with us. The server can be yours. The operations do not have to be.

An honest take on timeline

Standing up a container with a domain and SSL in an already standardized environment is fast. Fixing an AI-generated application that mixes development environment, keys in the code, and undeclared dependencies takes longer than the author expects. The work starts by making the repository shippable, not by “just pointing DNS.”