Before reading on, take a moment to look up “Moto Cães” by Thiago Bernardes de Souza, a second place winner in the 2025 iPhone Photography Awards, on the IPPAWARDS website.[1]
Then come back and read this.
Vibe coding produces rubbish. I’ve heard this argument from very capable people. So does traditional development, when discipline is absent.
They’re half right, and half protecting something.
The case against is familiar. Copying rubbish from the internet. Fragile, untested, poorly structured outputs. Serious development requires serious developers.
Not entirely wrong. Undisciplined AI-assisted development produces exactly those outcomes.
But it’s measuring the wrong thing.
That photograph was shot on an iPhone 15 Pro Max. The same device is in your pocket right now. The craft didn’t disappear when smartphones democratised photography. It just stopped being the entry requirement.
Discipline is the differentiator
Vibe coding is the same shift. AI has removed the barrier to entry for software development. That doesn’t make discipline optional. It makes it the differentiator.
Even Andrej Karpathy, the man who coined the term, has moved on. He now calls it “agentic engineering” and by late 2025 was writing that he could be ten times more powerful with better application of available tools, saying that a failure to claim the boost “feels decidedly like skill issue.”[2][3] The constraint, as he sees it, is how well people use the tools. Not the technology.
Y Combinator’s Winter 2025 batch makes the commercial case. Twenty-five per cent of startups built real companies on ninety-five per cent AI-generated code. Real revenue. Real venture backing. Ten engineers delivering what used to require fifty to a hundred.[4][5]
What we see in practice
We’ve put over a hundred executives through our Operationalising AI bootcamp building real solutions using tools including Claude Code, Lovable, n8n and Emergent. The quality challenge has never once been the issue. What we do see is people jumping ahead. One prompt. A credible-looking output appears and confidence takes over.
Then it unravels. Because they skipped the fundamentals. Define your requirements. Start with a thin slice, a single process, a single data source, a single user. Build that rigorously. Test it properly. Then extend. The people who struggle the most are often those who won’t slow down long enough to build on solid ground.
That’s not a vibe coding problem. That’s a methodology problem. And in my experience it’s the only problem worth solving.
Vibe coding doesn’t produce rubbish. Rushed, undisciplined, overconfident development does. It always has.
If you would like your team to build real AI solutions with the right discipline from the start, our education and training programmes may help. You can read more on our services page or get in touch.
This article was first shared as a post on LinkedIn.
Sources
- IPPAWARDS, Take a look at the winners of the 2025 iPhone Photography Awards (“Moto Cães”, Thiago Bernardes de Souza, second place, shot on iPhone 15 Pro Max).
- Andrej Karpathy on X, 26 December 2025 (“I have a sense that I could be 10X more powerful… a failure to claim the boost feels decidedly like skill issue”).
- Sequoia Capital / YouTube, Andrej Karpathy: From Vibe Coding to Agentic Engineering.
- TechCrunch, A quarter of startups in YC’s current cohort have codebases that are almost entirely AI-generated, 6 March 2025.
- CNBC, Y Combinator startups are fastest growing in fund history because of AI, 15 March 2025 (Garry Tan: “you don’t need a team of 50 or 100 engineers”).