One Month Without AI

Summary

The author recounts using AI coding tools at work while banning AI contributions from his open-source project, LibreWeddingPlanner. What began with autocomplete and generated functions expanded into coding agents that implemented Jira tickets, decomposed larger tasks, created commits and pull requests, and worked concurrently across multiple git worktrees. He says this made him lose track of which changes were necessary, while reviewing and correcting the agents’ output often took longer than doing focused work himself. The agents produced convincing but flawed code and tests, required repeated prompting, and once spent about $30 in tokens while stalled. A coworker later pointed out that one of the author’s tests did not test the changed application scenario, prompting him to stop using AI even at the risk of his job. After returning to test-driven development, smaller pull requests, concise descriptions, and human code reviews, he says he regained understanding, confidence, and enjoyment in programming. The essay argues that developers should examine their dependence on AI and preserve their own technical judgment rather than becoming passive supervisors of coding agents.