Planning software already tells a company what it will probably sell and what it should make. What it leaves to people is the decision that follows: when a part is short, who gets the stock, whether to pay for a rush delivery, and which customer waits. By 2030, some companies will let AI agents own narrow decisions like this inside strict limits, and a few already do it in pilots.
The weeks go to waiting, not to work
Picture a retailer with 40 stores and a winter coat. Six stores have sold out and four have too many. Sales wants the coats moved to the six. Finance does not want to pay for trucks on a low-margin item. Logistics wants full loads, and buying would rather reorder. Each person is being reasonable, and together they take one to three weeks to agree on a transfer. Most of that time is a request waiting in someone's inbox, not anyone's effort.
If the selling season is six weeks, a two-week delay gives away a third of it. That is the cost of the argument, and no better forecast would have removed it.
What an agent does differently
Planning tools could already run what-if scenarios. What they could not do was carry the answer to the other departments and choose. An agent that watches sales as they happen can see the gap in the six stores on the day it opens. It lists the options: move coats from the four stores, rush a reorder, or do nothing. It prices each one in money, counting the margin saved, the truck cost and the cash tied up in stock, and sends a person the ranked list with reasons. Where companies used to need several rounds of negotiation, the person now approves or changes a single recommendation.
The software vendors that run company records are already adding the pieces this needs, as SAP, Oracle and Workday have added agents to their systems.
The new work is writing down the exchange rates
An agent cannot tell that the sales director is the loudest person in the room. It needs to be told how many dollars a day of delay to the best customer costs, how much a rush truck is worth paying, and how much a pallet of idle stock costs to hold. Most companies have never written these numbers down. They live in the heads of the people at the table, and the answer changes with whoever argues best that week.
I expect this to be the first thing companies learn from agents, before any saving shows up. Two departments that thought they shared a goal will put different numbers on the same late order, and the gap between those numbers is the argument they have been having for years. The monthly meeting does not disappear, but it changes: instead of deciding each case, it reviews the rules and the few cases the agent could not settle.
Data does not have to be perfect first. Gaps in the records show up when a real decision is attempted, and that is the cheapest time to find them.
So pick one decision your team repeats, such as moving stock between locations or choosing who gets a short supply, and write down the three or four trade-offs behind it with a dollar figure on each. Even with no agent, the people involved will stop re-arguing the same case, and if an agent comes later, you will already have the brief it needs.