The Vibe-Coding Backlash: Why most criticism misses the point

Across LinkedIn and X, vibe coding is mocked as sloppy, insecure, or a threat to “real” software engineering, but much of that pushback reflects defensive posturing more than grounded analysis. Vibe coding, popularized in early 2025, is simply an AI‑assisted development workflow where you express intent in natural language and let an LLM generate, iterate, and debug code with you in the loop.​

What vibe coding actually is

Vibe coding describes a conversational, intent‑first way to build software: you describe goals, accept AI‑generated diffs, run and test, then iterate through natural‑language feedback loops until it works. The idea was popularized by Andrej Karpathy, who framed it as “giving in to the vibes” and letting AI handle the boilerplate while you steer outcomes, especially for low‑stakes prototypes and weekend projects.​

The irony on social media

Since the term took off, social feeds flipped from routinely critiquing junior code quality to suddenly romanticizing it, often to argue that AI tools are uniquely bad while juniors are uniquely careful, which doesn’t match most real‑world code reviews. The online debates are loud, but many of the strongest critiques come from folks who haven’t used these tools beyond cursory trials, which skews perceptions of their practical value.​

AI tools vs. junior baselines

Modern AI coding tools frequently outperform typical junior baselines on speed, structure, and consistency for common tasks, especially when guided by a developer who can review and test outputs effectively. Reviews of platforms like Bolt.new show mature scaffolding, sensible defaults, and strong project hygiene that reduce boilerplate while keeping codebases navigable and extensible.​

Bolt.new as a case study

Bolt.new’s V2 added an agentic mode that autonomously chains tasks like debugging and refactoring, while giving you control to approve or adjust steps, minimizing the “spray and pray” pattern critics worry about. Its guidance emphasizes safer handling of environment variables and secrets via dedicated stores or configuration patterns, reducing accidental leakage in shared repos or logs during rapid iteration.​

Prompting is requirements engineering

“Garbage in, garbage out” isn’t unique to AI, unclear requirements have always produced poor software; vibe coding just makes this more visible and faster to iterate. Treat prompts like lightweight requirements: specify constraints, acceptance criteria, context, and interfaces, then tighten the loop with tests and feedback for reliable outcomes.​

Use cases that shine

Vibe coding is most effective when speed and iteration matter more than heavy upfront design. It’s a natural fit for scripts and internal tools, where you can generate working code quickly, test it against real logs and data, and refine in tight loops. The same strengths carry over to prototypes and proof‑of‑concepts: you can validate feasibility without over‑engineering the early layers. As those ideas harden into MVPs, the workflow helps you ship core value fast, collect feedback, and progressively strengthen the codebase with tests and reviews. For high‑assurance systems with strict compliance or reliability requirements, pair AI assistance with rigorous code review, security scanning, and formal testing gates to uphold standards.

Security is a practice, not a promise

AI‑generated code can introduce known vulnerability classes if left unchecked, the same way hurried human code does; the remedy is standard security hygiene: dependency vetting, secret management, code review, and automated scanning. Organizations that operationalize AI coding alongside SAST/DAST and secret scanning cut the risk while keeping the productivity gains.​

The real takeaway

Vibe coding isn’t a gimmick or a replacement for engineering fundamentals, it’s an acceleration layer that rewards clear intent, tight feedback loops, and disciplined review. Teams that embrace it pragmatically ship faster prototypes, cleaner scaffolds, and safer defaults, while reserving deep review for the parts that truly matter.

Affiliate Disclosure

Some of the links on this blog are affiliate links, which means I may earn a small commission if you make a purchase through these links. This comes at no additional cost to you and helps support the content provided on this blog. Thank you for your support!