Ars reporter says intensive use of AI coding agents led to burnout after dozens of projects

Benj Edwards, Ars Technica's senior AI reporter, writes that extensive use of AI coding agents left him burned out after he produced dozens of hobby projects over a short period. In an opinion piece, he describes both the exhilaration and the exhaustion of pushing these tools hard.

Edwards says he used tools such as Claude Code and Claude Opus 4.5 via a personal Claude Max account, OpenAI's Codex, and Google’s Gemini CLI, paying out of pocket for premium plans. Fueled in part by a bout of COVID and a temporary 2x Claude usage cap, he completed more than 50 demo projects in about two months, including games and utilities such as a multiplayer Katamari Damacy clone called "Christmas Roll-Up," Card Miner, Violent Checkers, Flip-Lash, Grapheeti, and AI-enabled emulators for Apple II and Atari 800.

From that work he distilled 10 lessons: humans remain essential to design and long-term maintainability; models are brittle outside their training data; true novelty can be difficult to achieve; the first 90 percent of a project comes quickly while the last 10 percent is tedious; feature creep is seductive; and current systems are far from artificial general intelligence.

He also warns these tools can make creators busier, can produce work faster than audiences can absorb, and can amplify both productivity and pitfalls. Edwards argues AI coding agents should be treated as power tools that amplify human ideas rather than autonomous replacements.

Key Topics

Tech, Benj Edwards, Claude Code, Openai Codex, Google Gemini, Anthropic