Where Operators Come From
AI is removing the work that has long served as operations' training ground. So how do we build the next generation of experienced operators, and can we do it better than before?
By Kelly Breakstone Roth, Co-Founder & CEO of Prysmic · September 2026 · 5 min read
Every conversation about AI in operations lands, sooner or later, on the same question: if agents take over the junior work, where will the next generation of operators come from? It is usually asked by people who came up through that work themselves, and that is exactly why it deserves a serious answer. Ask any operator you trust where their judgment came from and you hear the same story: years of manual work, reconciling invoices, chasing carriers, re-keying figures between systems, until the patterns sank in. Nobody designed that education. The manual work was the school, and judgment was its byproduct. The work that built every senior operator you rely on is the work the agents are taking.
Leadership, meanwhile, is looking at the same shift through a different lens. In Gartner's survey of 509 supply chain leaders (published February 2026), 55% said they expect agentic AI to reduce the need to hire for entry-level positions. The logic is easy to follow: if the manual work is handled, the junior hire that existed to do it looks like a line you can strike. Gartner's own analysts predict the savings will not hold, and the reason is plain supply and demand. Pausing junior hiring does not pause the need for experienced operators three years from now; it pauses their development. And because much of the industry is pausing at once, the pool of early-career professionals with real experience shrinks for everyone at the same time. By 2030, Gartner expects 75% of the organizations that pause entry-level hiring this year to be paying premiums of 15% or more, bidding against each other for the few people who got the experience anyway.
The debate usually stops there, with one side protecting the junior hire and the other cutting it. Both sides share an assumption worth examining: that the old first job was a good way to learn. Look closely at how that education actually worked, and a third option appears, better than the one being defended and the one being cut.
How operators were actually made
Take the old first year apart, hour by hour, and most of it was moving information: downloading the report, pasting it into the model, fixing the lookup, sending the chasing email. Patricia Coan, who ran supply chains at L'Oréal, Tarte, and Jones Apparel and was most recently COO of the smart fragrance company Pura, did that work herself as a demand planner: "Just preparing the data could take a day or two, before you could even start the analysis." The learning lived in the other, smaller part of the job: the moments when something broke, a call had to be made, and the junior was close enough to watch it being made, or trusted enough to make it. Those moments built the instincts. Nobody could plan for them. They came with the crises, so a junior could go months without one, and two people with the same tenure could end up with very different educations.
For as long as operations has existed, judgment has been a byproduct. Now it has to be a product.
A step higher
Coan has already watched the first half of this shift from the hiring side. "I don't necessarily need my college entry-level analyst," she told us, "because the person who was doing that work is now these tools." The person coming in, she says, needs to come in "a step higher":
"More strategic, instead of spending the day on the robotic work of pulling data, inserting formulas, and sorting it all. That work has already been done. What I need from you is to tell me the decision, the direction we should be going in."
She does not mourn the old first job, either. Her own, in MRO purchasing, "was a lot of requisition processing, and I kept thinking, why did I go to college? This is paperwork." The new version, in her words: "more challenging and more engaging. But you have to bring critical thinking, and you have to be ready to actually decide."
She is describing the change one hiring decision at a time; PwC's 2026 AI Jobs Barometer sees the same thing across more than a billion job ads, where entry-level roles most exposed to AI are now seven times more likely to require traditionally senior-level human skills: leadership, creativity, face-to-face judgment. A step higher is the right description, and it carries an obligation most of the discussion skips. A junior who starts at decisions instead of data prep needs the decisions brought to them, with enough context to make them well. That does not happen by accident any more than the old education did. It has to be built.
The raw materials
An operation run by agents has the two raw materials the old first job never had. The first is the shape a decision arrives in. When an agent hands a call back, the case comes prepared: what happened, the evidence assembled, the history with that supplier or that carrier, the agent's own recommendation, every step traceable. Does the reorder justify air freight, or does the stockout cost less than the shipping? Is this supplier's third late confirmation a bad month or the start of a pattern? The old version of those moments arrived as a fire, and a junior spent most of it reconstructing what had happened while it was still happening. Now the reconstruction is waiting on arrival, and what is left to add is the judgment.
The second is the record. When agents do the work, every decision is written down as it happens: what was seen, what was decided, why. The moments that used to evaporate, the call a senior operator made at six in the evening and never thought to mention, now sit in a log a beginner can study and a manager can review with them, the way teams review film after a game. Tribal knowledge stops being a matter of which desk you sat at and becomes something you can actually teach.
Making operators on purpose
Building the education out of those materials is a management job. A new hire can spend their first weeks reading the record, studying how the operation actually decides, before they own a single call. Then the queue comes with an explicit boundary: decisions up to a threshold are theirs, everything above it escalates, and the boundary widens as their calls hold up, the same graded trust you would extend to anyone new to authority.
The overrides become the teaching material. When a senior operator overrides an agent's call, the override captures a rule that until then lived only in their head, and reviewing those calls with a new hire hands over tribal knowledge faster than waiting for the next crisis to teach it by accident. The role changes shape with it, measured in decisions made and how they held up rather than tasks closed, because the tasks left with the agents.
Coan offers the caution that belongs near the end of this argument: "make sure the tools can do everything you need them to do before you wholesale eliminate people," because "eliminating repetitive work doesn't mean eliminating people. It creates capacity. And leadership determines whether that capacity becomes cost savings or competitive advantage." The entry-level question is that choice in miniature. One company banks the salary, thins its bench, and five years from now is buying senior operators at a premium out of the shrinking pool everyone else's caution created. Another takes the capacity its agents freed and spends part of it running the education deliberately, and five years from now it is promoting people who came in a step higher and have three years of reviewed decisions behind them.
The debate will keep framing this as a tradeoff: adopt agents or develop people. Inside an agent-run operation, the tradeoff dissolves. The system that took the junior work is the same one that makes real development possible, every case arriving with its evidence, every decision on the record, every senior override written down where the next hire can learn from it. Meanwhile the companies hesitating for the pipeline's sake are holding their juniors in the old first job, learning by luck, while a competitor's new hires review real decisions with the evidence attached. Five years from now, when everyone is competing for experienced operators, where will the best of them have learned the job? It might just be inside the operations everyone feared would kill the pipeline.