Keywords
AI, Testing, Research, Coding, Research

Technologies
Github, TypeScript, SQLite, Next.js, Jest

Artificial intelligence coding tools are rapidly transforming how software is built, but a fundamental question remains: can they fully replace software engineers? This project investigates this by having four AI tools, ChatGPT, Claude Code, Cursor, and Firebase Studio, each independently build the same retail web application across 17 standardised tasks, evaluated using identical automated test suites.

Each tool was used according to its natural interaction model, some working conversationally, others self-testing and self-correcting between prompts. Results were measured across structural correctness, runtime behaviour, time taken, and human intervention required.

The findings suggest AI tools can produce functionally complete applications but differ in reliability and autonomy. The most capable tool completed all 17 tasks in under three hours with zero manual intervention, a pace hard for a human to match. The gap between passing tests and producing working software remained a persistent challenge. AI appears to be redefining the role rather than replacing the software engineer, shifting human effort from writing code to reviewing, directing, and correcting it.