A software developer chronicles a personal odyssey with AI coding assistants like Claude Code, Claude Opus 4.5, Codex, and Gemini CLI. What started as a thrilling exploration quickly evolved into burnout after tackling dozens of prototype projects in a short span, funded out of pocket and fueled by a curiosity that felt reminiscent of early programming days.
Like a 3D printer that creates flashy prototypes but struggles with durable production, today’s AI tools can spit out quick interfaces, games, and demos. They borrow patterns from training data, but turning a prototype into production-grade software requires architecture, discipline, and patience that the tools alone cannot supply.
AI models remain brittle outside their training data. They excel at patterns they’ve seen but stumble in unfamiliar domains. Production code demands robust testing, version control, incremental work, and human oversight to guide decisions and maintain long-term maintainability, something the author learned across more than 50 experiments.
Experts like Simon Willison have argued that AI tools amplify existing skill. The author emphasizes that AI is a tool, not a person or employee, and that experienced developers are still essential to steer projects, manage scope, and ensure quality beyond the tool’s immediate capabilities.
Despite the dazzling speed of AI, the author warns that these tools may increase, not decrease, workload in real-world software work. Mastery still takes time, and the best path is to harness AI to augment human judgment while keeping the human in the loop to guide architecture, decision-making, and creative vision.
