AI loops: who pays for the tokens?
Jack Franklin examines the "AI loop" problem: developers using AI coding tools spend significant time reviewing and correcting AI output, raising questions about who ultimately bears the cost of expensive token usage and whether the productivity gains justify the expense.
Background
Jack Franklin is a UK-based software engineer who wrote about the hidden costs of using AI coding assistants like Claude Code (Anthropic's tool for running Claude directly in the terminal). His core point: each time a developer asks an AI to debug, rewrite, or explain code, every round-trip burns tokens (the unit of billing for LLM APIs). The person who pays—the developer's employer or the developer themselves—often doesn't notice the accumulating cost of these iterative "loops," especially when the AI produces long, verbose responses. This matters because as AI coding tools become more common in professional engineering, the economics of token usage are still poorly understood: a single debugging session can cost more than the monthly subscription implies, and teams rarely track the per-session cost of asking "one more question."