OpenAI: 80% of enterprise AI problems are deployment, not models

OpenAI's head of forward deployed engineering Colin Jarvis said 80% of enterprise AI struggles stem from deployment, not model quality. Custom work per project has dropped from 90% to 50% thanks to tools like Codex, and one semiconductor project saves an estimated $40–50m a year.

• 80% of enterprise AI problems are about deployment, not models
• Custom work per OpenAI project fell from 90% to 50% with Codex
• One semiconductor project saves an estimated $40–50m a year
• One customer has about 35 AI use cases live after 18 months