A post on the Russian developer community Habr has a title that translates to "Vibe coding has completely won me over." But the very first line reads completely differently. The author, Vasjen, opens with "Everyone around me seems to have lost their minds lately." The title's "won me over" turns out to be pure irony. That gap between title and opening line is exactly what makes you keep reading. You expect another vibe-coding hype post, and instead you get the opposite argument.
Why the irony?
The author isn't angry at vibe coding itself, but at the hype surrounding it: people posting "I built a whole service with a few prompts," others selling courses off the back of it, a general vibe that software engineering as a profession has simply become unnecessary. He's careful to say this isn't about protecting his own job. Instead, he argues that any commercial system at real scale carries complexity no single person can hold in their head, complexity only a team that has worked together for years can actually understand. That's not something one prompt resolves.
Here he brings in Martin Fowler's Design Stamina Hypothesis: projects that invest in design start slow but keep their pace over time, while projects that skip design start fast but eventually hit a point where every new feature breaks an existing one. His diagnosis is that vibe coding traces exactly that second curve. Worth noting, though — Fowler himself calls this a hypothesis, not something backed by data.
The numbers back it up
So is this just an old-school developer grumbling? Recent data suggests otherwise. A 2026 survey found that misusing AI coding tools can actually slow work down by 19% and increase security vulnerabilities by 2.7x. Another survey found that 84% of developers have tried vibe coding, but only 29% said they trust the results. The pattern the author describes — fast in week one, then new features start breaking old ones around month three — is something that's now actually being observed. So when he welcomes vibe coding for dashboards, bots, and simple integrations but draws a hard line at CRM, CMS, and SaaS handling real user data, that's not an emotional reaction — it's closer to an observed pattern.
Where developer identity starts to wobble
What this post really touches isn't code quality — it's identity. To the question "if AI writes everything, what am I even doing here," the author redefines himself: not as someone who types prompts, but as the person who designs and takes responsibility for complexity. That redefinition hit close to home for me, because I've handed blog automation scripts to AI myself. Clearly scoped tasks — a publishing script, an image pipeline — got done in a few prompts, fast. But once several scripts started depending on each other, there were moments where neither the AI nor I could immediately tell what had broken or where. That boundary lines up exactly with the one the author draws: narrow, easily reversible tasks on one side, tasks where multiple parts depend on each other on the other.
Whether to use vibe coding as a tool or hand over an entire system comes down to who ends up carrying the complexity. I'm on the author's side here. I'll keep using AI aggressively for small, reversible tasks, but for anything where multiple parts interlock, I'm keeping the design in my own hands and treating AI as an assist, not the owner. The author has promised a follow-up with concrete failure cases — that one's worth watching for.
Top comments (0)