Amazon's $1.8 Million AI Experiment Gone Wrong: Cost Overruns Spark Concerns

Amazon's experiments with AI-powered coding tools have resulted in significant cost overruns, raising concerns about the company's ability to track and control its growing AI expenditures.

A recent project, codenamed Claude Sonnet, aimed at matching author details with product listings ballooned from an initial $207,000 budget to a final cost of $1.8 million - an increase of 860%. This overspending went undetected for nearly five months, sparking questions about Amazon's internal accounting practices.

This is not an isolated incident; multiple projects within the company have exceeded their budgets, including a financial auditing tool project that incurred an additional $541,000 in costs and a logistics-focused initiative with unplanned expenses of $134,000. The shift to token-based pricing models has made mistakes more costly, and AI agents' growing autonomy has allowed errors to multiply before human reviewers notice the damage.

Amazon acknowledges ongoing experimentation but disputes the characterization of these incidents as routine business practice. Executives across the industry continue to frame AI spending as a necessary investment despite growing evidence of inconsistent returns.