The problem with "vibe coding"

Usefulness for Experienced Developers

• Several experienced developers report strong gains from LLMs: faster design exploration (threads vs async vs procedural), better error-handling/logging, and rapid iteration on UX details.
• Consensus among many: tools are especially powerful in the hands of seniors who can steer, verify, and refactor; far less so for novices who can’t see when the model is wrong.
• Some describe LLMs as “tireless interns” that accelerate tedious work (boilerplate, enums, glue code) but still require expert oversight.

Learning, Expertise, and Hallucinations

• Debate over whether beginners can become real developers if the LLM does most of the coding.
• One side: you can inspect and ask for explanations, so you “get out what you put in”; LLMs can be better than static books for interactive clarification.
• Other side: hallucinations and shallow explanations risk teaching misconceptions; books and multiple human sources are still seen as more reliable foundations.

What “Vibe Coding” Is

• Working definition in the thread: having AI generate and iterate on most of the code while largely not reading it; copy–paste or agentic IDEs that autonomously edit, run, and fix.
• Some distinguish this from “LLM-assisted development,” where the human remains deeply involved in design and code review.
• There’s confusion and disagreement over whether exploratory use of LLMs (to test designs) is “vibe coding” or a different practice.

Programs vs Products, Maintainability

• Strong agreement with the article’s core point: a “works on my machine” program is very different from a maintainable, shippable product.
• Critics note vibe-coded output often looks polished but hides deep design flaws, concurrency issues, or brittle patterns that surface later.
• Concern that AI can generate so much code, in so many subtly different ways, that fixing systemic bugs becomes extremely hard.

Accessibility, Hacker Ethos, and Personal Tools

• A widely praised example: a non-professional using LLMs to build bespoke assistive software for a disabled relative—life-changing despite hacky, fragile code.
• Some call this the true “hacker spirit” (pragmatic, user-driven problem-solving); others argue traditional hacker ethos values deep understanding and elegance, not quick-and-dirty hacks.
• Many see LLMs as finally filling the old “HyperCard/Excel/VB” niche for end-user programming: personal tools tailored to one user, not commercial-grade products.

Jobs, Outsourcing, and the Developer Pipeline

• Speculation that LLMs may reduce the total number of developers or at least stall growth, while making remaining experts more valuable.
• Worry that, like the COBOL/mainframe world, we might end up with a shrinking pool of true experts and a broken pipeline for creating new ones.
• Some compare vibe coding to outsourcing: both can degrade quality when used naively, yet both can work when managed carefully.

Quality, Safety, and Long-Term Risks

• Strong skepticism about using vibe-coded systems in critical infrastructure (energy, finance, healthcare).
• Fears of a coming wave of poorly understood AI-generated code embedded in business workflows that degrades badly over a few years.
• Related anxiety about “empowerment/democratization” rhetoric being used to justify more surveillance, control, or cost-cutting at the expense of quality and labor.

Future of Practice

• Some envision prompts as the next programming layer, eventually with more formal, high-level description languages for AI.
• Others insist you can’t escape formalism: without structure, prompting is just “blind shots,” and software engineering discipline still matters.
• Overall divide: enthusiasts treat vibe coding as a powerful new RAD/glue paradigm; skeptics see it as throwing away hard-won engineering practices when building anything meant to last.