What is “vibe coding”?
The phrase vibe coding describes a workflow where a developer leans heavily on a large language model (LLM) to produce the bulk of a software project. Instead of writing functions line‑by‑line, the programmer supplies high‑level prompts and lets the model generate the implementation, tweaking only where necessary. The label was first used by Andrej Karpathy – the AI researcher who once headed Tesla’s Autopilot Vision team – in a February 2025 post on X, where he wrote, “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 that tweet, the approach has moved from a novelty to a mainstream practice, especially as commercial LLMs such as Cursor Composer with Sonnet have become capable of producing production‑ready snippets.
Why some developers love it
Advocates argue that AI‑driven code generation lowers the barrier to entry for software creation. A non‑technical user can describe a desired feature and receive a working prototype in minutes. The article cites a veterinary technician who built a simple app to log insulin doses for senior cats without any formal programming background – a classic illustration of the democratising promise.
For seasoned engineers, the appeal lies in speed. Routine boilerplate, repetitive API calls, and even some algorithmic scaffolding can be offloaded to the model, freeing developers to focus on architecture, performance tuning, and user experience. In a 2025 Stack Overflow developer survey, 47.1 % of respondents reported using AI tools every day, primarily for autocomplete, code review and quick problem‑solving.
The flip side: security and maintainability concerns
Critics warn that handing over large chunks of code to an opaque model can create hidden weaknesses. A study from Georgia Tech’s School of Cybersecurity and Privacy examined 43,000 security advisories issued over a three‑month window and traced 74 of them back to AI‑generated code. Fourteen of those were classified as critical vulnerabilities. While the absolute number sounds modest, the researchers estimate the true figure could be five to ten times higher because only disclosed, clearly‑marked AI‑generated snippets were counted.
Beyond outright bugs, there’s a more subtle risk: codebases that are heavily AI‑written may become difficult for human maintainers to understand. If a model‑produced function fails and the developer lacks deep knowledge of the underlying logic, fixing the issue can turn into a time‑consuming reverse‑engineering exercise.
How widespread is the practice?
A recent survey of 1,100 professional programmers who have experimented with AI tools revealed that 72 % now use such tools on a daily basis. On average, 42 % of the code they write is either fully AI‑generated or heavily assisted. Respondents also predict that AI‑generated code will make up more than half of their codebase within the next twelve months.
However, the same data shows a split in attitudes. While a majority employ AI for assistance, only a minority consider full‑on vibe coding part of their regular workflow. In the 2025 Stack Overflow survey, 72 % of developers said they did not incorporate vibe coding into their day‑to‑day work, and another 5 % said they were “emphatically” opposed to it.
Implications for the junior developer market
One of the less‑discussed side effects is the impact on entry‑level coding jobs. Historically, junior developers have been the workforce that handles repetitive tasks, giving them a foothold into the tech industry and a pathway to middle‑class salaries. As senior engineers increasingly rely on AI to automate those same chores, the demand for junior talent is shrinking. Companies are hiring fewer junior coders, and the skill set required to stay employable now includes prompt‑engineering and the ability to audit AI‑generated output.
What should developers do?
The consensus among experts is not to abandon AI tools, but to use them judiciously:
- Treat AI output as a draft – always review, test, and refactor generated code before merging it.
- Maintain clear documentation – note which sections were AI‑written so future maintainers know where to look.
- Invest in security audits – run static analysis and penetration testing on AI‑heavy modules.
- Upskill in prompt engineering – learning how to ask the right questions can dramatically improve the quality of generated code.
- Balance automation with mentorship – senior developers should still involve junior team members in code reviews to preserve learning opportunities.
Bottom line for Android‑focused developers and contract seekers
For Android app creators, vibe coding can accelerate prototype development, especially when building simple utilities or internal tools. Yet the security findings serve as a reminder that any AI‑generated component shipped to users’ devices must be rigorously vetted. As contract providers increasingly offer AI‑assisted development packages, buyers should ask providers how they validate and secure AI‑written code. A transparent workflow that combines rapid AI generation with human oversight will likely deliver the best mix of speed, cost‑effectiveness, and reliability.
The information above is based on Engadget’s coverage by Igor Bonifacic (Sept 6 2026) and publicly available research from Georgia Tech and Stack Overflow surveys.
