Vibe-coding tools ease setup but still require programmer expertise

A ZDNET writer tested AI "vibe coding" tools — Cursor, Replit, Microsoft Visual Studio with GitHub Copilot and Lovable — to see whether a novice could build a small data-analysis app using natural-language prompts. The author found the tools impressive at automating setup tasks such as creating a file structure, installing libraries, configuring a virtual environment and producing a basic web UI, and after several days managed a simple analysis program.

But each tool had notable limits: Cursor sometimes erased chat history stored in a workspace folder under ~Library with no way to retrieve it; Replit ran in the cloud and quickly exhausted free credits while failing to read some .Pages files; Visual Studio often required copying snippets into a terminal and produced only basic text-string matching; and Lovable, while fast and easy to use, could only get better results by routing text to Google Gemini, which due to .Pages parsing could analyze only titles and metadata.

The writer paid Lovable $25 per month for 100 credits and ultimately connected an XML archive on Algolia so Gemini could perform rudimentary semantic analysis, producing a basic viable product. The experiment exposed trade-offs between desktop and cloud offerings: desktop tools posed stability and workflow headaches, while cloud services ran up costs and required uploading private data.

Key Topics

Tech, Cursor, Replit, Visual Studio, Lovable, Google Gemini