Usage is not value.
I’ve been building more AI into real workflows lately, and one thing keeps standing out — most internal AI metrics don’t tell you whether anything actually improved.
We track token usage. We track tool adoption. We count POCs.
But none of that answers the question that actually matters:
Is the business better because of this?
Did quality go up? Did cost go down? Did delivery speed up? Did error rates drop? Did the people doing the work get real leverage?
I’ve been thinking about this more because the recent examples are hard to ignore.
McDonald’s ended its IBM AI drive-thru test after repeated ordering problems. Uber reportedly burned through its 2026 AI coding budget in just four months. Microsoft is pulling back internal Claude Code licenses and pushing teams toward Copilot CLI.
To me, these aren’t “AI is failing” stories.
They’re what happens when adoption moves faster than governance.
And I think that’s the real gap.
Not more models. Not more tools.
A real operating layer around the AI teams already have — governance, observability, cost controls, evaluation, clear ownership, and a way to roll things back.
We should probably stop treating internal AI leaderboards like scoreboards.
Usage is not value.
We should be a lot more skeptical of AI POCs too.
If a POC has no path to production, no measurable return, no budget guardrails, and no one who actually owns it, it’s not strategy.
It’s an experiment with a nicer slide deck.
The teams that get this right won’t be the ones using the most AI.
They’ll be the ones disciplined enough to keep asking:
What is this actually doing for us? What is it costing? Can we prove either one?
I’m still figuring out what the right balance looks like in practice.
But the pattern keeps showing up:
Fast adoption without real governance gets expensive faster than most teams expect.
#AIGovernance #GenAI #LLMOps #AI
