Agentic coding is a financial trap: $12 to fix a single off-by-one error
The era of predictable $20-a-month AI subscriptions is ending, replaced by autonomous coding agents that burn through pay-per-token API budgets. One engineer spent over $12 for 45 minutes of automated React refactoring, and a typical agent resolving a standard ticket can cost nearly $12 by re-reading 120,000 tokens across six loops. The author argues that deterministic, manually-scoped tools with local models are far more cost-effective for maintaining complex codebases.
Bloated code is a feature, not a bug. It spends more tokens creating, spends more tokens refactoring, and makes it harder for a human to take over, thus locking you to the AI.