Two complaints about AI are being repeated across the industry right now:
"We're not seeing the productivity gains."
"We're spending way too much on tokens."
They look like two problems. They're the same one.
Both come from the same well-meaning instruction: "Apply AI everywhere."
Spread AI evenly across every task, every workflow, and the result is predictable: AI costs scattered across work that doesn't need it, and gains too diluted to show up in business results.
The mandate feels like leadership. It's actually the absence of it - a decision about where AI belongs, delegated to nobody, enforced everywhere.
AI is just an evenly distributed expense.
The biggest drain on my own org was never building new things. It was sustaining what we'd already shipped: bug triage, regressions, the escalations that eat into the team's week.
So that's where we pointed AI. Not everywhere, but there ... and the hours came back.
The harder, more valuable work is choosing where AI actually belongs: the workflows that move revenue, margins, speed, or customer outcomes.
That judgment can't be handed to a blanket mandate. It requires leaders who understand both the business and the technology well enough to see where the payoff is.
Target those workflows deliberately. Aim the tokens where the value is.
Then both complaints fade at once: returns climb because AI is aimed at work that matters. Costs fall because you're no longer paying for work that doesn't.
It's AI where it counts.