Remember Doing the Work
In robotic surgery, residents spent 10 to 20 times less time on surgical tasks than in open surgery.

At the second console, the resident watches the screen. The senior surgeon sits about 15 feet from the patient, controlling the robot. Whether the resident gets control is up to the surgeon. The surgeon owns what happens to the patient, so the handoff is a decision about how much of the live case to actually let the resident do. The surgeon still judges, in the moment, whether the resident's hands are good enough to keep working on the patient, and takes the controls back the moment they aren't. Keep every rep, and the resident gets no real practice, but hand it over, and the surgeon is the one who decided the patient was safe in those hands.
Matt Beane studied this arrangement at five hospitals from 2013 to 2015. He observed 94 procedures over 478 hours. Sometimes the resident was at the bedside instead, holding a suction tool.
The procedures averaged 4.5 hours. In open surgery, residents worked with their hands for essentially the whole procedure. In robotic surgery, they spent 10 to 20 times less time on surgical tasks.
Beane's words for the resident's place in the operation were "an optional participant."
The resident was there to learn. But the operation no longer required the resident to do the work. The surgeon could keep going while the resident watched.
If you have been asked to review work you never learned to do, that second console may feel familiar.
Where the learning went
The approved route was simple: watch the work, do the work, then teach it. The robot loosened the link between those steps.
The robot loosened the link between those steps.

Click a numbered dot to see what it stands for. Start with 1.
The residents who made exceptional progress found ways around the approved route. Some specialized early, at the expense of broader training. They used simulators and watched recorded surgery on YouTube.
They also worked near the edge of their ability with little expert supervision.
At the five hospitals, Beane found only three residents who had made exceptional progress. In a separate interview study, attending surgeons rated 12 of 33 residents exceptional. Nine were average or below.
The residents who had not become exceptional could still finish training authorized to perform robotic surgery.
A senior surgeon Beane interviewed put it plainly: "Watching a movie doesn't make you an actor."
Yet getting control could mean losing it again, in public. One resident described being left "like a kid sitting in the corner with a dunce cap."
Beane studied learning, not patient outcomes.
His findings do not establish patient harm. They show a training problem: the residents who learned found practice that the normal arrangement no longer reliably provided.
Who learns to check?
Supervision has to come from somewhere.
In 1983, Lisanne Bainbridge wrote about automation in process industries. She described physical skills that "deteriorate when they are not used." The knowledge needed for an unusual situation, she wrote, "develops only through use and feedback about its effectiveness."
The unusual situation matters most. When an automated process needs manual help, something may already be wrong. The person taking over may need more skill than ordinary work requires.
Bainbridge also saw a problem across generations. Former manual operators could monitor automated systems using skills built before automation. Later operators could not be expected to arrive with those same skills.
I read that as a ladder with a missing rung: later operators are asked to watch and judge without first learning the work by hand. She wrote that forty years before AI at work.
Later operators are asked to watch and judge without first learning the work by hand.

Click a numbered dot to see what it stands for. Start with 1.
My concern is that experienced people can hide the gap.
My concern is that experienced people can hide the gap.

Click a numbered dot to see what it stands for. Start with 1.
They still catch the error because they remember doing the work.
Their success tells us little about whether anyone behind them is learning how.
The people coming next
Does that concern reach beyond an operating room?
Stanford researchers Erik Brynjolfsson, Bharat Chandar and Ruyu Chen examined ADP payroll data. Their August 2026 revision covers employment through June 2026.
Workers aged 22 to 25 in AI-exposed occupations stood 19% below where employment would have been had it kept pace with less-exposed peers. That is a relative shortfall, not a count of jobs proved lost to AI.
The original 2025 paper reported a different pattern for more experienced workers in the same occupations. Their employment had stayed steady or grown.
Young software developers showed a sharper fall. By July 2025, their employment was nearly 20% below its late-2022 peak.
The researchers also distinguished automation from assistance. They found young-worker employment falling where AI primarily automated work, rather than where it most helped people do it.
For AI-assisted work, a review can pass honestly without showing who can do the work. The residents' training could authorize them to operate without proving they could operate alone.
When the numbers mislead
The employment evidence is unsettled. CNBC reported in August 2025 that the working paper had not been peer reviewed. The authors caution that factors other than generative AI may play a part.
The Yale Budget Lab reached a different result in October 2025. Using different data and methods, it found no clear relationship between AI exposure and employment changes. Its young-graduate findings could also reflect a slowing labor market.
That matters. A smaller intake does not prove automation caused it. But I would still worry about fewer people getting practice, whatever closed the door.
Bainbridge's last irony ran the same way: the most successful automated systems, she wrote, "may need the greatest investment in human operator training."
For AI at work, I expect the first cost at the second console. Nobody on the team can trace an error back to the person only watching. That is an expectation, not a finding.
If cohorts hired now supervise as well as those before them, I would be wrong.
Back at the second console, the surgeon can finish the operation.
The next surgeon is still waiting for a turn.
I'd like to hear what the people coming next on your team get to do first.
Part of The First Move, on why judgment depends on who moves first, and on who decides what reaches the customer.