Wolfgang Rohde paper warns AI labor substitution erodes long-term capability

In a new paper on arXiv (arXiv:2605.27399), researcher Wolfgang Rohde challenges the prevailing narrative that AI labor substitution is a pure efficiency win. He introduces two key mechanisms: 'capability masking' and 'capability erosion.' Capability masking occurs when AI-generated output makes it appear that an organization's human expertise has been replaced, when in reality skilled labor is still needed to verify, refine, and maintain that output. This illusion drives hiring freezes and budget cuts, allowing the underlying human capabilities to slowly degrade. Rohde supports his argument with evidence from AI-assisted coding tools, showing that generated code often suffers from correctness bugs, poor maintainability, and security vulnerabilities that only experienced developers can catch. Repository-level studies further reveal that AI struggles to understand full codebase context, limiting its ability to handle complex, production-grade systems.

Beyond individual developer productivity, Rohde examines broader labor-market and political-economy forces. He notes that managerial cost incentives and national competition for AI supremacy push organizations toward rapid substitution, increasing risks of vendor lock-in and market concentration. The paper synthesizes findings from software engineering research, industrial strategy, and political economy to paint a sobering picture: what looks like acceleration may actually be a quiet transfer of risk from the present to the future. Organizations that cut human expertise too aggressively may find themselves without the foundational knowledge needed to troubleshoot, innovate, or adapt when AI tools fail or reach their limits. The result is a system that appears more efficient in the short term but becomes structurally fragile over time.