Relearning and team formation for clinicians in the age of generative AI: steps and practice for bringing perspectives together
Contents
- AI in the team’s daily work
- AI that keeps working on your behalf
- What each person’s AI illuminates
- Bringing independent judgments together
- Students and residents with always-on AI
- Four steps in a fictional discharge conference
- Five elements for structuring AI answers
- When to look at AI
- Practicing how to convey
- Reviewing your own strengths
- References
AI can now say accurate things. How those things are conveyed, however, still depends on people. That is why the value of a team lies in each person bringing the abilities they have enhanced with AI.
The research behind this claim and its connection to Buber’s I and Thou are covered in “All real living is meeting”: Teams in an age when AI answers accurately. This article focuses on how to bring those abilities together across professions and where to practice doing so.
AI in the team’s daily work
AI is already part of the team’s daily work. In the United States, a clinician-facing AI assistant called Art is being built into Epic’s electronic health record1. At NYU Langone Health, clinicians can use the AI search tool OpenEvidence inside the Epic record with the patient’s context2.
In Japan as well, a workflow is spreading in which generative AI creates summaries from conference recordings, and a person checks and corrects them before they go into the record. At Aizawa Hospital, the hospital and the vendor reported that the time to write records of pre-discharge conferences fell from 60 to 10 minutes, and the number of conferences rose from 4 per month before introduction to up to 20 per month3.
So, if knowledge is available on the spot, does judgment get better? In a randomized trial of 50 physicians, there was no statistically significant difference in diagnostic reasoning scores between the group that could use GPT-4 and the group with conventional resources only (details of the trial).
Simply making AI available did not change the quality of judgment. I think the difference lies in how the team discusses and decides. How to convey and how to decide remain on the human side.
AI that keeps working on your behalf
In autumn 2026, AI that keeps working on a person’s behalf also appeared4–9: Claude Duet (Anthropic) patrols connected apps every few hours and sends a cross-app summary every morning; Google’s coworker agent has its own email address and appears in the employee directory; xAI’s Team Bots are role-specific bots shared by a team and can be invited to Slack channels; OpenAI’s dots has its own computer and browser in the cloud and keeps working even when your devices are off; and Meta’s Muse keeps working after the app is closed and asks for approval before sending email or making purchases.
Then, is it enough to add one agent to the team?
What each person’s AI illuminates
AI works like a magnifying glass or binoculars, letting us see what we are interested in more deeply and farther away. What lies outside our interest, on the other hand, tends to get less attention (the magnifying-glass analogy). A physician’s AI deepens the physician’s view, and a nurse’s AI deepens the nurse’s view. Unless these are brought together, the patient’s wishes and the differences between professions are handed to no one’s AI.
AI does not join the team as a member; it is a tool that sharpens each person’s eyes.
Bringing independent judgments together
Another point to watch is that a wider view for each individual and less diversity of perspectives within the team can happen at the same time. Stories written with AI ideas were rated higher, while the stories became more similar to one another (the story-writing experiment).
The key to keeping a team diverse is bringing independent judgments together. Reading together a study that combined diagnoses made separately by physicians and AI, and a meta-analysis of human–AI combinations, combinations work when each judgment is independent and there is a procedure for deciding which to adopt (the two studies).
Students and residents with always-on AI
Students and residents come to clinical settings with always-on AI. The question for education shifts from whether to use AI to how the team makes up for what each person’s AI does not see.
Four steps in a fictional discharge conference
As a procedure for relearning across professions, I propose four steps (the list of four steps). Here I follow them through a fictional pre-discharge conference for a patient who wants to return home.
Step 1: Write down observations before looking at AI
Each profession briefly writes down its own observations and concerns: what they saw, what they do not know, and what they would prioritize. The nurse writes “unsteady at night,” the pharmacist “sleeping pill can be continued,” the therapist “step at the entrance,” and the physician “consider a care facility.”
Step 2: Supplement with AI
Ask AI for evidence, alternative hypotheses, and overlooked points, and list them in a form that shows they came from AI. In this example, “fall-prevention guideline” appears as evidence, “effect of the sleeping pill” as an alternative hypothesis, and “family’s caregiving capacity” as an overlooked point.
Step 3: Examine differences and gaps
Check why opinions differ, which generated content lacks a basis, and where proposals conflict with the patient’s wishes or living conditions. The pharmacist’s “sleeping pill can be continued” conflicts with the AI’s “effect of the sleeping pill.” The source of the fall-prevention guideline cited by AI cannot be confirmed. The physician’s “consider a care facility” does not match the patient’s wish to go home.
Step 4: Decide together and record it
Decide together, and record the reasons for adopting or rejecting each proposal and who is responsible. Record the reasons for rejection as well. In this example, reviewing the sleeping pill is adopted (hypothesis about the unsteadiness; pharmacist), the fall-prevention guideline is rejected (source could not be confirmed; physician), and a meeting with the family is adopted (consistent with the wish to go home; nurse).
Five elements for structuring AI answers
Adding AI does not change the framework for decision-making. Applying EBM and finding common ground in patient-centered medicine remain as they are. I am considering structuring the AI answers that enter this framework with five elements: (1) a structured question (PICO), (2) a recommendation (direction and strength), (3) a summary of the evidence and its certainty, (4) statements that support decision-making, and (5) transparency about references and sources.
The fourth element, statements that support decision-making, does not stop at “consider the patient’s values” but shows where in the balance of benefits and harms the choice would change. Even so, the decision belongs to the patient and the team.
When to look at AI
The reason for not looking at AI in Step 1 is that if everyone looks at AI first, their judgments drift toward the AI, and the value of bringing them together decreases. If each person’s observations come out first, the differences remain. In interprofessional education in Hong Kong, students also made care plans without AI, checked them with AI, verified them against guidelines and expertise, and then approved them jointly, in this order (the Hong Kong example).
I think the task for education is less about teaching how to use AI than about making sure that different observations and dissenting views remain in the team after AI is used.
Practicing how to convey
How to convey things can be practiced. As a place for that, I have created and published a smoking-cessation clinic simulator (in Japanese). When you choose responses as the physician, a fictional patient answers in voice, and feedback appears at the end.
What this simulator lets you experience is that even medically correct information can fail to reach the other person depending on how it is conveyed. Trying to correct someone with correct information only pushes them to look for reasons to keep smoking and to reassure themselves, reinforcing that rationalization.
Running a team has the same structure. Even when the right direction is clear, it will not reach the patient unless the team decides how the professions will approach them together.
Reviewing your own strengths
Other professions are also using AI. When the team discusses a case, it is now possible to get high-resolution advice from different angles. So what can you bring?
AI can also be used to review your own strengths. Ask AI about the following three things and put them into words: the aspects of the patient your profession sees, the changes you notice first, and the judgments you take responsibility for.
Nurses observe daily life and change, pharmacists see how medications are actually taken, therapists look at movement and environment, and physicians integrate diagnosis and treatment plans. The clearer each role becomes, the wider the view when they are put together (each profession’s view).
What we bring to the team is not AI itself but the abilities each of us has enhanced with AI. The wide-angle view is held by the team. This, I believe, is how teams should be formed in the age of generative AI.
What remains in the end is facing one another in the same place. What Buber’s “All real living is meeting” asks of this age is discussed in Reading “encounter” in the age of AI.
References
- A Deep Dive Into the Announcements at Epic UGM 2025. Healthcare IT Today. 2025-08-21. https://www.healthcareittoday.com/2025/08/21/a-deep-dive-into-the-announcements-at-epic-ugm-2025/
- OpenEvidence. Announcements. 2026-10-29. https://www.openevidence.com/announcements
- Ubie Inc. Announcement on the introduction at Aizawa Hospital (in Japanese; PR TIMES). 2026-07-01. https://prtimes.jp/main/html/rd/p/000000216.000048083.html
- Anthropic. Claude Duet. 2026-10-21. https://claude.com/resources/articles/claude-duet
- Google Cloud. Welcome to Gemini at Work 2026. 2026-10-09. https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026
- xAI. Team Bots. 2026-09-28. https://x.ai/news/team-bots
- OpenAI. Meet dots. ChatGPT Learn. https://learn.chatgpt.com/docs/dots
- Meta. Introducing Muse: The World’s First Personal AI Agent Built for Everyone. 2026-09. https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/
- Meta. Muse personal AI agent launches in Japan (in Japanese). 2026-10-23. https://about.fb.com/ja/news/2026/10/muse-personal-ai-agent-launches-in-japan/