Some uncomfortable truths about AI coding agents

This post argues against using LLM-based AI coding agents for production code, citing four core concerns: skill atrophy, artificially low cost, prompt injections, and copyright/licensing.

• •Engineers shifted to code review roles will gradually lose hands-on coding skills as they stop writing code themselves
• •Reviewing AI-generated code at scale leads to inevitable complacency, allowing bad changes to slip through
• •Generative AI models are wildly unprofitable; current pricing is disconnected from actual training and inference costs by orders of magnitude
• •A bubble pop would force massive price hikes and usage caps, making AI-dependent workflows suddenly unviable
• •The author recommends avoiding over-dependence on AI coding agents before the market correction arrives