Incentives for AI use: A ‘spectacularly bad idea’

Summary

This article addresses the growing organizational trend of using quotas and incentives to drive employee adoption of AI tools, arguing that such approaches are fundamentally flawed. The author contends that top-down, individually measured AI adoption mandates produce superficial compliance rather than genuine productivity gains, as employees either fabricate usage metrics or apply AI tools unnecessarily to meet quotas. Key evidence includes the self-reporting problem inherent in most AI usage tracking, the observation that most current AI use amounts to substituting AI-powered search for traditional search, and a cited figure suggesting only approximately 5% of AI projects produce meaningful organizational results. The author draws an analogy to lean production principles, arguing that group-level, discretionary, employee-driven experimentation yields more authentic improvement than individually mandated targets. The article concludes that effective AI integration requires psychological safety, visible peer-led examples, group recognition rather than individual quotas, and transparency about job security to eliminate employees' rational disincentives to participate in their own potential displacement.

Strengths and limitations

The article presents a logically consistent critique of incentive-based AI adoption strategies, and the core concern about Goodhart's Law — that measuring a proxy metric corrupts the underlying behavior — is well-established in management literature. The lean production analogy provides a recognizable reference point for group-based discretionary improvement. However, the article is entirely opinion-driven with minimal empirical support; the single cited data point (5% AI project success rate) is attributed only to a 'State of AI in Business study' without full citation details, making it unverifiable. The author provides no data on outcomes from the group-based, discretionary approaches they advocate as alternatives. The argument conflates different types of AI incentive programs without distinguishing between poorly designed self-report quotas and more sophisticated outcome-linked measurement systems. The anecdotal observation about programmers checking AI-generated code is presented as representative without sourcing. The article reflects a clear normative stance against managerial control mechanisms, which may introduce bias in how counterarguments are weighted.

What this implies

The article points to a tension between organizations' desire for rapid, measurable AI adoption and employees' rational responses to poorly designed measurement systems. The cited 5% project success figure, if accurate, implies that current PMS-linked AI adoption strategies are not translating into organizational value. The critique of individual-level metrics in favor of group-level, outcome-based recognition suggests a potential directional shift in how AI integration is embedded within performance frameworks. The programmer example raises broader questions about whether productivity-linked performance metrics will adequately capture changes in work quality versus work volume as AI augments task execution.

Key points

• Individual AI usage quotas and incentives are likely to produce metric compliance rather than genuine productivity improvement, as self-reported measures are easily gamed and difficult to verify.
• Approximately 5% of AI projects are reported to produce real organizational results, suggesting that individual-level mandates are insufficient substitutes for structural organizational change.
• The assumption that AI will handle routine tasks while elevating employees to higher-value work is challenged by evidence from software development, where AI-generated code has shifted human work toward more tedious error-checking rather than more meaningful tasks.

What to take away

• Organizations measuring AI adoption through self-reported individual quotas face a fundamental verification problem, making group-level, outcome-oriented approaches a more observable alternative.
• Psychological safety and transparent communication about job security are identified in the article as preconditions for voluntary, genuine employee engagement with AI tools.