My 2027 prediction for the future of work (Thinkers360)
Thinkers360 published my 2027 prediction on the future of work. AI supervision should be part of the job, and healthcare already has a model for teaching that judgment.


Thinkers360 asked a group of us for our 2027 predictions on the future of work. Their piece, Thinkers360 Predictions Series: 2027 Predictions for Future of Work, was published on October 18, 2026. I was one of 13 contributors.
Here is the prediction I submitted, as published on the Thinkers360 page:
In 2027, I expect employers using AI agents to make supervision of those systems an explicit part of ordinary job descriptions. Hiring and promotion will increasingly test whether someone can spot a plausible but wrong answer, explain the error and decide when to stop an automated workflow. Healthcare offers a useful model: give people responsibility in stages, with supervision and demonstrated competence before greater independence. As AI absorbs routine tasks, employers will also need to preserve the practice that helps junior staff develop the judgment required to oversee it. The organizations that do this well will measure productivity after checking quality and accounting for the time people spend correcting AI mistakes.
Harvey Castro, MD, MBA, Emergency Physician and Healthcare AI Speaker | Founder at DR GPT
In the emergency department, the chart still needs a name on it. If a tool drafts a note, flags a patient, or suggests a next step, someone has to notice when the answer looks right and is not, explain why, and stop the workflow if needed. That skill should be written into the job, not treated as a side habit.
For clinicians and health leaders, I would take two things from this. First, supervision of AI should be an ordinary part of the role, with staged responsibility, the same way we already train clinicians. You do not hand someone full independence on day one. You expand the scope after they show they can do the work. Second, productivity numbers that ignore quality, and the time spent fixing AI mistakes, will flatter the tool and hide the real cost.
There is also a training problem in the efficiency story. If AI absorbs the routine work that used to teach junior staff how to think, we still have to give them practice. Otherwise we will ask people to oversee systems they never learned to question.
This is the same tension I took to 41 leaders for The State of AI in Emergency Medicine 2027. When ED AI gets it wrong, who answers? Those answers are contributor perspectives, not a representative survey.
My AI 2027 predictions report launches at CES on January 8, 2027. I will have more to say there about judgment, supervision, and accountability.
This post is educational. It is not individual medical, legal, or employment advice.
