TL;DR
Vibe coding is real: describe what you want and an agent hands back a working app, no engineering background required. Where nothing depends on it and nothing is lost if it is wrong, that is the right way to build it. A business-critical application is a different discipline, with different people, a different setup and different requirements. Agents made it easier for almost anyone to build software, and harder to find engineers who can build the business-critical kind. The hype is built on the demo, because "built in three days" sells and careful engineering has no headline. Before the work starts, name the mode: vibe coding or agentic engineering.
Decide what happens to the code before you decide how to build it. If the business will come to depend on that code, it takes agentic engineering. For a demo, vibe coding is enough. Both use the same tools, so only that decision tells you which one you need.
Vibe coding fits wherever a wrong result costs nothing.
#Say what you want and an returns something that runs, with no engineering background needed. That part of the promise turned out to be true, and there is no reason to be sour about it. Where nothing will come to depend on it and a mistake costs nothing to undo, is the right instrument for the job: a quick prototype, an internal utility that cannot damage anything, a static website.
The test is not how long it lives. A script that runs once against the production database lives for seconds, but a mistake in it can destroy data nobody gets back. Vibe coding is the wrong instrument for it.
The obligations arrive with the owner, not with the technology.
#Enterprises have already run this experiment. Business-critical functionality gets built in a spreadsheet by the people closest to the problem, and for a while it works.
Its author moves on, nobody upgrades it, nobody tests it, and whatever is quietly wrong inside it stays wrong, because nothing was ever built to catch it.
The spreadsheet did not fail on the day it was written. It failed on the day the business started depending on it and nobody owned it.
The spreadsheet was quick to build, but on credit. It was paid for later, with interest. A vibe-coded app the business comes to rely on does the same, only faster. Ownership and review rules cover software someone has declared business-critical. Nobody ever declared the spreadsheet business-critical; it simply became so. A vibe-coded app can quietly become business-critical the same way.
Anyone can vibe code; only a few can do agentic engineering.
#The hype has this exactly backwards. Vibe coding removed the entrance requirement, so the number of people who can produce something that runs is effectively unbounded. moved the other way: it asks for an engineer who can already build business-critical software and has also learned to direct agents well. Building that kind of software means judging structure, security and what a system will cost to own. Before agents, building business-critical software was enough. Now it takes that and directing agents well, and far fewer engineers can do both. And the work that needs them did not become any less critical.
What sells is the calendar time, never the engineering that made it safe.
#I have often seen AI sold to a business with a vibe-coding demo. Such demos wow the audience, because so much comes together in so little time. But what vibe coding produces is built for the moment, and a lot of work is still needed before a business can rely on it. "We built it in three days" is a sentence anyone can repeat and everyone can picture. The engineering that makes a system safe to own has no such sentence, and the demo never mentions what it costs. So the three days get quoted, and that cost stays out of the pitch.
Neither mode is a mistake; using one for the other's work is.
#Nobody pitches a tent expecting to live in it for twenty years, and nobody calls that a failure of tents. A decision, a budget or a direction may rest on a prototype, and it is still a prototype.
Keep exploring