The cost of being half-hearted in AI and how to avoid the Solow Paradox

The article discusses the challenges faced by enterprises in adopting artificial intelligence (AI), drawing parallels to a phenomenon known as the Solow Paradox. This paradox was first observed by Nobel laureate economist Robert Solow in 1987, who noted that although computers were prevalent in innovative businesses, productivity statistics did not improve.

Several companies, including Uber, have found themselves in a similar situation. For Uber, the issue stemmed from blowing through its entire 2026 AI budget within four months, largely due to around 5,000 engineers relying on Anthropic's Claude Code. This situation is not unique, as Forrester research found that enterprises are deferring approximately 25% of their planned AI spend to 2027, as CFO scrutiny over return on investment (ROI) intensifies.

The article highlights that while 62% of companies are experimenting with AI agents, only 23% have scaled them in a single business function. Uber's president and COO, Andrew Macdonald, raised concerns about the rising cost of AI token usage not translating into proportional productivity gains. Uber's approach to AI adoption, which involved ranking teams by tool usage with the incentive to use more tools, led to higher usage but made the business impact harder to justify.

This demonstrates the importance of redesigning workflows alongside the adoption of AI tools, rather than simply optimizing their usage.

The article also explores the distinction between AI activity and AI maturity. Running numerous pilots, adopting multiple platforms, and reporting impressive usage statistics does not guarantee measurable business value. Uber's experience serves as a cautionary tale, emphasizing the need to focus on business outcomes rather than merely tool usage metrics.

When AI token usage is unconstrained, activity becomes the proxy for progress; however, when usage is capped, organizations are forced to confront the more critical question of what each token is actually producing. This economic mirror reveals the gap between activity and productivity, where most AI ROI is lost.

The article suggests that individual task optimization, while beneficial, is not sufficient for AI maturity. Instead, enterprises should focus on production-ready AI, which has four key characteristics. Firstly, the output must feed directly into downstream decisions or actions without requiring manual transfer. Secondly, the system should have clearly defined failure modes, ensuring accountability and remediation.

Thirdly, there should be a named owner responsible for performance, adoption, and iteration. Lastly, and crucially, the surrounding process should be redesigned, not merely augmented.