Series 02 — Vibe coding
Vibe coding is fine until the demo has a URL.
Two people, the same agent, the same morning. Both leave with a demo that looks finished. Three weeks later, only one of them is still shipping.
01 / The scene
The other one is still paying for it
By the afternoon both had something that looked finished, and both were right to feel good about it.
Three weeks on, one of them is still shipping. The other is paying interest. The shortcuts that made the demo fast are now the work that has to be redone, and the thing that could not take traffic is the thing that has to. Same model, same day, and, from the outside, the same prompts.
One of them, on purpose or by accident, noticed the moment the demo stopped being a demo. The other never said it out loud.
Turn your prototype into a structured engineering workflow without losing the work already done. Book a 30-minute pilot call to explore how Woky can help.
02 / The definition
It is the same cycle, with the stops turned off
Vibe coding is not the absence of a way of working. It is the same cycle, with the approvals in the middle switched off and a single look at the end. Context, structure and the files are still there. You just do not see them while you type.
What gets removed are the moments when a person confirms scope, architecture and plan before anyone starts. What remains is one look at the result. The agent still reads the files, follows the conventions and leaves something behind, without stopping to ask. When that works, the structure was already good enough, or nobody was going to depend on what came out.
The useful argument is not whether the mode is real. It is where the line sits, and who is allowed to move it.
03 / The failure test
It works because a wrong turn costs minutes
At demo stage there is no user, no data, and often no URL. If the output is wrong, you delete it and ask again. The cost is minutes, not an incident. Speed matters more than the trail.
I have done this in front of someone who does not write software. There was no project, no repository, no plan. There was a conversation. I was showing the product and said we could make landing pages the same way. Then I did it, live, while we kept talking. Five minutes to create the project and its generic example page. Six to generate the English page from a PDF of a product presentation. Three to translate it and fix it until the Spanish version was there too. At minute fifteen both versions existed, and we had not stopped talking.
Then I showed the source and the terminal. Under the demo was a real project, not a screenshot. The speed was not a trick laid on top of nothing. It was a real build, with the approvals taken off for a while.
04 / The boundary
Three questions, and then it is no longer a demo
A mock is disposable until it has a URL, users, or data. Any one of the three changes what the thing is. A URL means someone can find it. Users mean someone depends on it. Data means someone can lose something.
Past that line, vibe coding is the wrong mode. Not because the code suddenly turned bad, but because one look at the end no longer covers the risk. The failure is not technical. Someone never said, out loud, that the demo had stopped being a demo.
Apply it: does it have a URL? Users? Data? If any answer is yes, you are not prototyping anymore, even if the folder still says prototype.
That night, it already had a URL. It had crossed the line before anyone named the crossing. The demo was still fine. The way we were working was not.
05 / The switch
The next morning, the page stayed
Approvals come back. So does a record of what was decided, and a review before anything ships. Scope is confirmed before the work; the result is approved before it goes out. Between those two moments, each step leaves the work and the reason, so the next person can read the decision instead of rebuilding it from the chat.
Changing the cycle does not mean rewriting the project. The next morning I read the result without rushing, decided it was good, and moved it to production. I took off the agile cycle that project did not need, set up the cycles it uses now, gave it the roles it needed, and took the domain.
The page was never rewritten. It is the same site from that fifteen-minute session, with a production cycle on top of it. You are reading this on that site. What changed was the cycle, not the code.
06 / The protection
The protection is a person with a job, not a checklist
A cycle does not protect you because it has steps. It protects you because each step has a role, a written mandate, and a verdict someone can inspect. The roles below are not people, and they are not autonomous agents. Each one is a responsibility. If a role has no failure behind it, it does not belong in the table.
| Role | Protects against |
|---|---|
| Product Owner | Requirements with no acceptance criteria |
| Architect | Structural decisions nobody recorded |
| Feature Planner | Ambiguous tasks that each person interprets differently |
| Project Manager | Work without order, closure or memory |
| Frontend Developer | Execution without a code standard |
| UX Developer | Pages that are usable but have no design criteria |
| App Tester | Changes nobody verified |
| SEO Engineer | Publishing without being findable |
| AI Search Engineer | Publishing without being citable by an AI |
| Marketing & Sales Validator | Publishing without anyone understanding what is sold |
Two human approvals sit above them. One when the scope is confirmed. One when the result is approved. Nothing else stops a person. Everything between those two runs without asking. That is where the speed comes from.
Woky coordinates the implementation workflows. OrkAI maintains the project knowledge, requirements, decisions and standards behind them.
07 / The objection
A stop is justified by a scar
Nobody adds an SEO reviewer for the pleasure of it. They add one the week they published a page nobody could find. They do not add an AI Search Engineer because the title sounds current. They add one the day an assistant recommended a competitor and cited nothing from the site.
If you cannot trace a stop to a real failure, it is ceremony. Delete it. If you can, it is cheaper than the incident that taught you to add it. That is the only good reason to keep a ceremony, and the test that keeps a cycle from turning into bureaucracy.
08 / The risk
The bill does not arrive as a ticket
The expensive mistake is not prototyping in vibe coding. It is leaving the prototype cycle on a product that is already in production, because nobody said the work had changed. That cycle has no stops in the middle, no trail, and no review. On something people use, that is not speed. It is risk nobody is holding.
The bill arrives as an incident. Said early, changing the cycle costs an afternoon. Found late, it costs a rewrite, a migration, or a customer. The three questions are cheaper than the rewrite. Every time.
Vibe coding is the right way to work when being wrong costs nothing. The moment the thing has a URL, users, or data, change the cycle, not the code.
“This is bureaucracy with a nicer name.”
No. Bureaucracy is a stop with no failure behind it. Every role above answers a specific way the work has already gone wrong. If you cannot name the failure, delete the stop. That is a budget, not a process.
“If the prototype works, why touch it?”
Because “works” describes the conditions, not the code. A prototype that works for one person, on one machine, with nothing to lose, is doing a different job from a product with a URL. The code can stay. The cycle has to change.
“Nobody can tell when a prototype becomes a product.”
Someone can. The person who notices that it now has a URL, users, or data. Three questions, each with a yes or a no. The moment is knowable. The failure is that nobody wrote it down.
Ready to take your prototype further?
Start with a 30-minute conversation and explore the right way to evaluate Woky.
Book a 30-minute pilot call