Survey: 82% of enterprise AI spend lost to bugs, rewrites

Entelligence AI surveyed 2,444 companies and reported that up to 82% of enterprise AI engineering spending is absorbed before software reaches production. The firm allocated token-related costs as follows: for every $1 spent on AI tokens, $0.44 goes to bug fixes, $0.27 to rewrites of AI-generated code and $0.11 to review and merge delays.

Lightrun’s 2026 State of AI-Powered Engineering Report found 43% of AI-generated code still requires manual debugging in production after passing quality checks. The report also found no engineering leader surveyed expressed full confidence in deployed AI output.

Oracle has accumulated about $108 billion in total debt and raised roughly $50 billion in 2026 through debt and equity to fund AI data center expansion. The company’s free cash flow was near negative $13 billion. More than $300 billion of a $553 billion backlog is tied to work for OpenAI, which reported an approximately $14 billion loss last year. Oracle’s June 16 earnings report will be watched for signals about demand for its AI capacity.

OKX has tied employee evaluations to AI proficiency as part of new performance rules for the exchange. CEO Stax Xu argued that AI agents can speed execution while exposing employees who prioritize impression management over measurable outcomes. Xu added the AI era changes talent requirements.

The reports show companies are paying for model access and AI tools while engineering teams spend time and budget making automated outputs reliable and production-ready. Coinbase’s phased AI rollout and Cardano’s split approach to AI-generated code are cited as examples of similar quality-control and deployment challenges.

Investors and engineering leaders are monitoring debt levels, upcoming earnings reports and defect metrics to track whether current spending patterns will result in reliable, production-ready software and sustainable returns.