Kate Holterhoff explores the "vibe coding" phenomenon that defined the summer of 2025, a casual and intuitive approach to AI-assisted software development. "Vibe coding," a term popularized by Andrej Karpathy, involves letting large language models (LLMs) do most of the coding work, while the developer guides the process in natural language, without meticulously reviewing every line of generated code.
Simon Willison highlights the potential of "vibe coding" for learning and experimentation, presenting it as a democratizing force capable of enabling millions of people to build their own custom tools without requiring a computer science degree. However, not everyone shares this enthusiasm. Shawn "swyx" Wang warns against the "negligence" that "vibe coding" could encourage, where developers stop at a "good enough" level without completing the difficult 20% of work involved in polishing and fixing bugs. He stresses the importance of efficiency and quality, suggesting a more rigorous approach such as AI engineering.
The summer of 2025 saw a proliferation of tools and features supporting "vibe coding." Improved LLMs such as DeepSeek-v3, Gemini, Claude, Grok-4, and ChatGPT made "vibe coding" more accessible. IDEs such as Cursor and Windsurf, as well as SaaS platforms such as Lovable, v0, Bolt, and Replit, also emerged. AWS even integrated a "Vibe" mode into its Kiro IDE, allowing AI agents to generate code and infrastructure with minimal instructions. GitHub launched Spark, a platform that turns ideas into intelligent full-stack applications with a single click.
Despite the enthusiasm, major challenges emerged, particularly around security. Incidents such as the loss of a production database at Replit or data leaks via the Model Context Protocol (MCP) at GitHub and Supabase forced providers to strengthen their security measures. Austin Parker of Honeycomb advises "verifying the results" of AI-generated code and applying DevOps principles of observability and testing.
Another point of contention was Cursor's pricing policy, which irritated the "vibe coding" community due to confusing price changes and unexpected costs for heavy use of AI models. This incident highlighted the economic reality of "vibe coding": heavy token usage is expensive, and providers cannot offer unlimited access indefinitely.
The article also mentions the emergence of "agentic" coding tools, where AI can autonomously break down high-level tasks into coding subtasks. Cognition's (the creators of Devin) acquisition of Windsurf illustrates the strategic importance and financial stakes of this technology.
In conclusion, the "Hot Vibe Code Summer" was a period of experimentation and enjoyment, making coding more accessible and intuitive. While security and cost issues remain, the spirit of "vibe coding" is likely to endure, integrating its key characteristics into standard development practices.