In-House AI Coding Raises Token and Upkeep Risks

Generative AI tools including Anthropic’s Claude and OpenAI’s ChatGPT have lowered barriers to software development, encouraging companies to build applications in-house and reduce reliance on outside vendors. But usage-based token charges are only part of the bill. Security reviews, regulatory compliance, system integration and ongoing maintenance can erode projected savings, especially when AI-generated code must be monitored and revised as underlying models change.

Industry advisers are warning that companies routinely underestimate those post-launch costs. Token spending can rise sharply as usage expands, while model updates may break workflows and require continued engineering oversight and human correction. More than 40% of agentic AI projects are expected to be abandoned by 2027 as escalating costs, unclear business value and inadequate risk controls outweigh anticipated productivity gains.