Startup founder says one Claude setting drove a $30,000 AI-token month

Business Insider reports that Turbo AI cofounder Sarthak Dhawan said his small team accidentally spent about $30,000 on Claude Code tokens in a heavy shipping month. He said the cost came partly from leaving fast mode on, and the team now uses fast mode only when latency matters. The piece is a concrete operating example of AI-agent costs moving from abstract token economics into startup budgets and engineering habits.

Key details: Turbo AI spent about $30,000 on Claude Code tokens in one month, The founder linked part of the bill to leaving fast mode enabled, The team now reserves fast mode for latency-sensitive pairing.

Why it matters: AI coding tools are becoming material operating expenses, and teams are learning that model settings can matter as much as vendor pricing.