Large U.S. companies are putting more money into AI agents, and their chief executives mostly describe the results as useful, not transformational. About 39% report significant gains in efficiency or growth, 43% report moderate gains from targeted uses, and 9% call agents transformational. That spread is what you would expect from companies that have added AI to the steps they already had.
The old layout hides a rule nobody remembers
The steam-era factory was not badly designed. It was shaped by one fact: power came from a single point, so everything had to stay close to it. Electric motors removed that fact, because every machine could have its own motor. But the floors, the belts and the order of the machines stayed, and for years owners saw an electric factory that looked like a steam factory and worked like one. It took a generation to notice that the rule behind the layout was gone.
Offices carry the same kind of hidden rule: a person has to read each thing. Approval chains, review queues and handoffs between teams exist because someone had to open the document, check it and pass it on. AI removes that rule, and the process built around it is still there.
What redrawing looks like in an invoice queue
Take a mid-sized distributor that receives 1,500 supplier invoices a month. A clerk spends about five minutes on each, typing it in and matching it to the purchase order, which is roughly 125 hours a month. Adding AI to the typing might save half of that. That is the electric motor in the old building.
Now redraw it. The AI reads every invoice the day it arrives and matches it to the purchase order and the delivery note. Invoices that match are scheduled for payment with no one touching them. Perhaps one in ten does not match, and those reach a person with the mismatch already explained, about 150 cases at ten minutes each, or 25 hours. The clerk's month drops by about 100 hours, and that is the smaller gain. Because invoices are checked on arrival, the company can pay fast enough to take the common supplier terms of 2% off for payment within ten days. On $1 million of invoices that qualify, that is $20,000 a month.
The second benefit never appears if AI is only used to type faster.
The first rung of the career ladder is the first thing to move
Three in four CEOs say AI has made them rethink the skills they want in entry-level roles. The invoice example shows why. Typing and matching were the work a newcomer did to learn the business. When the queue is redrawn, a new hire starts with the exceptions, which means starting with judgment about suppliers, prices and what looks wrong.
It also explains how 53% of these CEOs can plan cuts in some areas while 89% expect more people overall. Redrawn work shrinks some jobs and creates others, and the people in them are not the same people or doing the same tasks.
The person who knows the work has to draw the new one
A chief executive cannot redraw an invoice queue. The clerk who knows which suppliers always send mismatched totals can. That is why it matters that most employees are never told why their company uses AI: the people best placed to see which steps exist only because someone had to read the page are the ones left out of the plan.
For your own team, a good place to start is one list: which steps in this process exist only because a person had to read something? Those are the steps the new layout will not need.