J.P. Morgan: Generative AI Must Boost IT Workloads by 55% to Offset Cost Deflation

Generative AI is reshaping IT services, but it must drive over 55% more workload to offset significant cost deflation, according to a recent J.P. Morgan report. The brokerage estimates that large language models (LLMs) can cut project costs by approximately 35% through automation, especially in routine coding tasks—where deflation could reach 90%. These tasks currently account for 40% of full-time equivalent (FTE) hours in typical IT projects.

The report also highlights cost compression in areas like documentation and testing, though roles requiring complex reasoning—such as requirement analysis and architectural design—are expected to remain relatively stable. A notable shift in project structure is anticipated: code writing will drop from 40% of project costs to just 6% in the post-AI era. Meanwhile, tasks like prompt engineering, debugging, and code review will claim larger shares.

Prompt engineering alone is projected to represent 23% of post-AI project composition, up from 15% of pre-AI costs. Code review is also expected to double in importance, driven by increased scrutiny of AI-generated code. Debugging is forecast to remain cost-neutral, as generative AI can both streamline and complicate the process.

J.P. Morgan’s model shows that to maintain revenue levels amid a 35% price drop, IT services must boost workload volume or complexity by at least 55%. Encouragingly, early signs suggest this is achievable. Firms like Genpact (NYSE:G) have already reported a 3% revenue lift from clients adopting AI-based agentic solutions, indicating reinvestment trends similar to past tech shifts such as cloud adoption and offshore delivery.

As AI adoption accelerates, the structure of IT labor is evolving—not shrinking—pointing to a redistribution of roles rather than widespread job elimination.