In the early 1900s, Frederick Winslow Taylor became one of the most influential people in American business.

Taylor believed something radical for his time: work could be measured.

Factories largely operated on intuition. Managers relied on experience, workers developed their own methods, and nobody really knew where time was being spent or why productivity varied so dramatically between teams. Taylor thought that was a problem worth solving.

Armed with a stopwatch, he began measuring everything. How long did it take to move a piece of steel? How many shovelfuls of coal could a worker move in an hour? Which motions created waste and which created output?

The goal wasn't simply measurement. It was visibility.

Once managers could see where effort was being spent, they could finally make informed decisions. Before measurement, every productivity problem looked the same. After measurement, bottlenecks, inefficiencies, and high-performing processes became obvious.

More than a century later, many companies find themselves facing a similar problem with AI.

They know what they're spending, but they don't know who's generating it.

That distinction is starting to matter. Earlier this month, reports surfaced that Uber burned through its entire 2026 AI coding budget in just four months. The story wasn't really about Claude Code, Cursor, or any particular tool. It was about visibility. When usage grows faster than accountability, costs become a surprise instead of a decision.

This pattern is emerging across nearly every industry. Engineering teams adopt Cursor. Claude Code gets rolled out. GitHub Copilot is already running. Internal agents begin making API calls around the clock. AI moves from experiment to operating expense while finance receives a single invoice at the end of the month.

The most common mistake is routing every AI request through the same organizational API key. The simplicity feels great at first: one account, one invoice, one place to manage billing. Then every team disappears into the same cost bucket.

Maybe Platform Engineering is responsible for most of the spend. Maybe an internal agent is generating thousands of requests each day. Maybe you're paying for multiple coding tools and nobody can tell which one creates the most value.

Without attribution, nobody knows where the money is going. Without that visibility, optimization becomes nearly impossible.

That challenge is becoming more important as AI budgets grow. Gartner found CFOs entering 2026 balancing increased AI investment with growing pressure to control costs and demonstrate measurable returns. That's difficult to do when every dollar looks identical on the invoice.

Fortunately, solving the problem is surprisingly straightforward. Start by provisioning AI access per team instead of per company. Every team should have its own API keys. Every internal agent should have its own credentials. Every environment should map back to an owner.

That single decision creates the attribution layer most AI invoices will never provide.

Once you know where usage comes from, you can start asking better questions. Which teams are generating value? Which tools justify their cost? Which agents should be optimized? Which experiments deserve additional investment?

Taylor's stopwatch wasn't valuable because it measured time. It was valuable because it revealed where time was actually going.

Most companies don't have an AI spending problem yet. They have a visibility problem. And like the factory managers of Taylor's era, the companies that can see where resources are being consumed will make better decisions than the companies that simply pay the invoice every month.