In 2025, people “vibe code” by asking Claude, Cursor/o3, and Comet to build and modify software end-to-end—pushing PRs without reading the code, one-shotting games and web apps, wiring databases via MongoDB’s MCP, and even generating thousands of projects (one team reports $48,952.95 in model spend with zero human-written code). The tradeoffs show up quickly: posts describe production friction and outdated dependencies, loss of code understanding and fast-accumulating tech debt, and a security failure where a vibe-coded app exposed user data via a simple GET; practical mitigations include asking the agent to clean, de-bloat, and document after it finishes. Tools keep lowering the bar (Gemini 2.5 Pro for coding; OSS agent platforms that read/write files and run commands), but the guidance is to use vibe coding for prototypes or constrained features and apply human review before scale.
Been using Claude 3.7 for a couple days and it is very, very good.
Its "vibe coding" from language is impressive. I wrote a whole Substack post about the model (& Grok 3), link in reply.
Here is a one-shot prompted video game based on the Melville story “Bartleby the Scrivner” https://t.co/D8YkTR34oT