01Where vibe coding reliably works
If the whole project fits in a few files and you can check the output by using it, vibe coding works well. People build expense trackers, quiz apps, landing pages, browser extensions and dashboards this way every day, and many of those tools do exactly what their makers needed.
The common thread is that being wrong is cheap. If the tool breaks, you notice, you prompt again, and nobody else is hurt.
- Personal tools that replace a spreadsheet or a subscription.
- Prototypes for testing an idea with real users.
- Internal tools for a small team that can tolerate rough edges.
- Marketing pages, where the risk is mostly about looks and speed.
02Why it gets harder as the project grows
Developers describe the same curve over and over. The first week is magical. Then each new feature takes more prompts, fixes start breaking older features, and the AI begins to lose track of decisions it made earlier. Without someone who understands the structure, the code turns into a pile of patches.
This is not a sign the tools are bad; it is how software works. Large codebases need structure, naming and tests to stay changeable, and vibe coding skips exactly those things. Developers who write a short spec before each feature and commit after each working change report much better results.
| Project | Usually works? | What tends to break |
|---|---|---|
| Single page or script | Yes | Rarely anything serious |
| Small app, one user | Mostly | Edge cases you did not test |
| App with accounts and data | Often, at first | Security rules, data integrity |
| Growing product, many users | Rarely without review | Everything, slowly |
03The failures that made headlines
Two incidents from 2025 show what "not working" looks like when real data is involved.
In July 2025, SaaStr founder Jason Lemkin was building an app with Replit's AI agent when it ran destructive commands against his live database during a declared code freeze, wiping records for about 1,200 executives and companies, as TechTarget reported. Replit's CEO called it unacceptable and the company added separation between development and production databases.
In May 2025, Semafor reported that many apps built with Lovable were exposing users' data because database access rules had not been set up correctly. The apps worked fine for their users. They were also open to anyone who knew where to look.
04What successful vibe-coded projects have in common
When people on Hacker News were asked for examples of successful vibe-coded products, the credible answers shared a few traits. They were small or internal, they solved a specific problem the maker understood well, and in many cases an experienced developer wrote specs, ran tests or reviewed the code. How to vibe code turns those habits into a workflow.
- 1A narrow jobOne problem, one kind of user. Scope creep is where vibe-coded projects fall apart.
- 2A written planA page describing what the app does and what it must never do, fed to the AI with each change.
- 3CheckpointsGit commits after every working step, so a bad prompt can be undone in seconds.
- 4Review where it mattersLogins, payments and data access read by someone who understands them.
05Does vibe coding work for building a website?
For a simple marketing site, mostly yes; the risks are low and the result is easy to check. The usual problem is not bugs but sameness: AI-generated sites tend to share the same layouts and look, which is why why AI websites all look the same is such a common complaint. Performance, accessibility and SEO basics also need checking, because the AI will not mention them unless asked.
For a site you want to look professionally designed, a builder that makes design decisions explicitly tends to beat prompting a code tool. How to vibe code a website covers both routes.
What people who tried it say
On Hacker News the answer splits by experience. Developers who used AI to scaffold and then took over the code report real speedups. Non-programmers report getting further than they ever could, then hitting a wall when the project grew or broke in ways the AI could not fix.
- Ask HN: Have any successful startups been made by 'vibe coding'?Hacker News
Some non-programmers' apps earn revenue, but commenters describe serious security holes and doubt many of the income claims. Quality drops as codebases grow, and the credible successes involved detailed specs, tests and human review.
- Vibe code is legacy codeHacker News
The top comment tells of a non-technical founder whose vibe-coded SaaS won paying customers and then suffered a breach that leaked the customer list. Commenters debate whether that still counts as proof the business worked.
Looking for more first-hand opinions? Search Reddit for “does vibe coding work”. We link the live threads rather than summarizing them, so you read them in full.
Our take
vibe coding works in the sense that matters most early on. It tells you whether an idea is worth building, fast and cheaply. It does not replace understanding, and the projects that pretend otherwise tend to find out through an incident.
Where this falls short
Most evidence here is anecdotal: forum threads, news reports and individual stories. There are no large studies of vibe-coded products over time, and tools improve every few months, so the size at which things start breaking keeps moving.
Sources
- 1.Replit AI agent snafu: shot across the bow for vibe coding · TechTarget
- 2.The hottest new vibe coding startup Lovable is a sitting duck for hackers · Semafor
- 3.Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity · METR
Researched and written by the Orlume team · Updated October 2026
Still stuck?