Vibe coding draws fans and security warnings as AI writes more software

Karpathy introduced the term on X in a February 2025 post: "There's a new kind of coding I call 'vibe coding,' where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." Since then, the report says, large language models have become better at generating code, and the popularity of vibe coding has grown significantly.

The report describes both sides of the debate. Enthusiasts argue that vibe coding democratizes software development; Engadget said a neighbor who had worked as a veterinary technician used it to build an app for tracking her senior cat's insulin shots. Critics contend that AI-generated code can introduce vulnerabilities and result in codebases that are difficult to maintain, especially when a user does not know enough programming to fix a problem the model cannot resolve.

There is preliminary evidence for the security concerns. In April, researchers at Georgia Tech's School of Cybersecurity and Privacy scanned 43,000 security advisories and found 74 vulnerabilities that could be traced directly back to AI-generated code, including 14 they judged critical. They estimated that the actual number could be five to 10 times higher because they could only trace code explicitly disclosed as AI-generated.

The report says the practice now extends beyond hobbyists. A survey of 1,100 professional programmers who have tried AI tools found that 72 percent used AI coding tools daily and that about 42 percent of their codebase was AI-generated or AI-assisted. The same group predicted AI-generated code would make up more than half of their codebase by next year.

Engadget distinguishes between code written entirely by an LLM and code where AI assists with cleanup, debugging or suggestions. Stack Overflow's 2025 developer survey found that 47.1 percent of respondents used AI tools daily, but 72 percent said vibe coding was not part of their workflow and another 5 percent said it was "emphatically" not how they do their job. Most professionals, the report says, were using AI for autocomplete, code review and chatting through roadblocks.

The spread of AI also has implications for entry-level programmers. Senior developers now automate many tasks they once delegated to junior team members, and companies are hiring fewer junior coders. The report notes that coding used to be one of the few reliable routes into the middle class for young people, a path that is no longer as open.