AI Doesn’t Have ROI

Source: Where’s Your Ed At Date: 2026-06-02 URL: https://www.wheresyoured.at/ai-doesnt-have-roi/

Summary

Ed Zitron’s newsletter post argues that enterprise AI spending has not produced measurable economic returns after four years and trillions of dollars invested. The piece cites a Bain & Company survey finding 40% of companies saw AI-driven improvements of 10% or less, Uber’s CTO admitting the company burned its entire annual token budget in four months, and one unnamed company accidentally spending $500 million in a single month on API calls. The central claim is that subscription pricing deliberately obscured real costs to drive adoption, and that the absence of clear, defensible ROI stories is structural rather than temporary.

Implications

• Capital markets / AI value-chain skepticism thread. This piece is a data-dense articulation of a position gaining traction in enterprise finance circles: AI spend is real, but verifiable productivity gains are not. As token-based billing replaces flat subscriptions, cost visibility will force the ROI question into the open.
• Coding-agent competition thread. Developer tooling (Copilot, Codex, Claude Code) is the product category with the clearest productivity narrative — but also the one where Copilot’s 50%-credits-in-one-prompt episode most directly illustrates the cost-opacity critique. Vendors that can produce auditable productivity metrics have a differentiator here.
• Agent orchestration. Uncontrolled token burn at Uber and the $500M accident both point at missing spend-limit primitives in agent pipelines — a gap that is being actively addressed in Claude Code and competing platforms, but unevenly.