Why Enterprise Engineering Still Struggles to Prove AI ROI
SD Times, Thursday, September 17th, 2026
Token counts and lines of code both failed as proxies for value, pushing teams toward business-unit-level measurement.
As generative AI adoption accelerated through late 2025 and into 2026, CFOs began pressing engineering leaders for hard evidence that soaring token spend delivered business value.
Tempo.io CTO Shams Chauthani describes spend going through the roof as they adopted the tools, with the CFO asking what they were getting for it.
Organizations first relied on raw usage metrics such as token consumption, but high volume proved a poor proxy for productivity.
The industry then moved to output metrics like lines of code or pull requests, tracked in tools such as Atlassian DX and Jellyfish, which gave managers visibility into activity but did not answer executive questions about value, and penalized developers working on technical debt.