AI and Radiology Residency: What Program Directors Should Keep in Mind

What You'll Learn

This guide teaches one core skill you can apply in many different ways: how to collaborate with AI as a thinking partner, not just a task executor. By the end, you'll understand:

  • The difference between using AI as a tool vs. collaborating with AI
  • Nine specific techniques across three levels of collaboration
  • How to move from "co-thinking" (conversation) to "co-working" (autonomous tasks)
  • Practical how-to guides for Claude, ChatGPT, and Google AI tools
Key Principle: You Are the Expert

AI is powerful, but you bring irreplaceable expertise: clinical judgment, institutional knowledge, relationships with residents, and understanding of your program's unique context. The goal isn't to hand over decisions to AI—it's to amplify your effectiveness. You remain the "human in the loop" (a term popularized by Ethan Mollick) who guides, validates, and makes final calls.

From Tool to Collaborator

The way we think about AI is shifting fundamentally. Understanding this shift is the foundation for everything else.

Old Paradigm: AI as Tool
You give AI a task. It produces output. Transaction complete. Like using a calculator or search engine—helpful but passive. You do all the thinking; AI just executes.
New Paradigm: AI as Collaborator
AI participates in your thinking process. It plans with you, challenges your ideas, surfaces connections you missed, and helps you refine your approach through dialogue. The relationship is iterative, not transactional.
Strategic Thought Partner

A concept articulated by Jeremy Utley (Stanford d.school): AI functions best not as an answer machine but as a thinking companion. A strategic thought partner engages with you across every phase of a project—planning, research, discussion, creation, and feedback—helping you think better rather than thinking for you.

Co-Thinking

Co-thinking is interactive collaboration—you and AI working together through conversation. The three levels below (Capture, Challenge, Coach) represent increasing depth of collaboration, from you leading the process to AI guiding it.

You Lead
CAPTURE
You drive the conversation. AI assists with capture, storage, and transformation.

At this level, you're firmly in control. You have ideas, tasks, or information—AI helps you capture, organize, and transform them. This is where most people start, and it's immediately useful.

1. Voice Dictation

Speak your thoughts naturally and let AI transcribe, organize, and respond.

Why it matters: When you speak instead of type, ideas flow more freely. Typing naturally triggers self-editing—you stop to fix phrasing, second-guess word choices, and lose momentum. Dictating bypasses that filter, letting your thoughts come out in a raw, uninterrupted stream that AI can then help you organize and refine.

2. Second Brain

Create a persistent knowledge repository that AI can reference across conversations.

Upload documents, past decisions, policies, and context—then have AI help you make connections and retrieve relevant information.

What is a "Second Brain"?

A second brain is an external system that stores your knowledge, ideas, and reference materials in a way that's searchable and connectable. With AI, this becomes interactive—you can ask questions of your accumulated knowledge, not just search it. The term was popularized by Tiago Forte.

For Program Directors: Imagine uploading your program's policies, past CCC notes, accreditation documents, and curriculum outlines. Then asking: "Given our history with milestone assessments, what should we consider when implementing the new EPA framework?"

3. Output Formats

Transform your thinking into different formats without starting from scratch each time.

The same core content can become an email, slide deck, handout, FAQ, or even a script. AI handles the reformatting while you focus on the substance.

Co-Lead
CHALLENGE
You and AI think together. AI pushes back, surfaces alternatives, and expands perspectives.

At this level, AI becomes an active thinking partner. Instead of just capturing and transforming your ideas, it challenges them, offers alternatives, and helps you see blind spots.

1. Non-Obvious Solutions

Explicitly ask AI to go beyond the obvious and explore less traveled paths.

When you present a problem, AI's first suggestion is often conventional. Push it to explore alternatives you haven't considered.

Example Prompt "I need to improve resident engagement in our didactic sessions. Before suggesting solutions, first list 3 obvious approaches I've probably already considered. Then give me 3 non-obvious or counterintuitive approaches that might work better."

Why this works: AI has absorbed countless approaches to common problems. By asking it to distinguish obvious from non-obvious, you filter for insights that are actually new to you.

2. Sparring Partner

Ask AI to argue against your position to stress-test your thinking.

This isn't about AI being right—it's about stress-testing your thinking before you commit to a decision or communicate it to others.

Example Prompt "I'm planning to implement a new moonlighting policy. Here's my draft: [policy]. Play devil's advocate. What are the strongest arguments against this approach? What objections will faculty and residents raise?"

3. Expand the Team

Introduce AI into team discussions and departmental workflows to multiply impact.

AI collaboration works best when shared with colleagues.

How to use this: Bring AI into your next faculty meeting or committee discussion. Use it live to capture ideas, synthesize different viewpoints, or draft action items in real time. Share a project workspace with colleagues so everyone can contribute context and benefit from the same AI-assisted analysis.

Key benefit: When teams use AI together, they build shared mental models for what works. One person's prompting technique becomes everyone's. Collective learning accelerates individual adoption.

AI Leads
COACH
AI guides the process. You provide input while AI structures the thinking.

At this advanced level, you flip the dynamic: instead of you driving the conversation, AI guides you through a structured process. This is powerful when you're stuck, unclear on what you need, or facing a complex problem.

1. Planning Mode

Let AI structure a project before you dive into execution.

Describe what you want to accomplish; AI creates a plan, timeline, and identifies potential obstacles.

Example Prompt "I need to prepare our program for the ACGME site visit in 6 months. Don't start working on anything yet. First, create a comprehensive project plan: what needs to be done, in what order, who should be involved, and what the key milestones are. Then ask me clarifying questions before we proceed."

The power of "don't start yet": By explicitly asking AI to plan before executing, you get a roadmap you can review and adjust rather than outputs you have to redo.

2. Iterative Research

Have AI self-evaluate its output and improve through testing and iteration.

Instead of accepting AI's first output, create an iterative improvement cycle where AI reviews its own work, identifies gaps, and refines the result. This mirrors the scientific method: generate, test, improve, repeat.

How this works: After AI produces initial research or analysis, ask it to read what it created, identify what questions or concerns the output raises, gather additional information to address those gaps, and then incorporate the new insights into an improved version.

Example Prompt "Create a research briefing on competency-based medical education models for radiology residency. Then: (1) Read your own briefing as if you were a skeptical program director. (2) List 3-5 additional questions this briefing raises that aren't fully addressed. (3) Research answers to those questions. (4) Integrate those insights into an improved version of the briefing."

Why this matters: AI's first attempt is often surface-level. By building in self-evaluation and iteration, you get deeper, more nuanced outputs that anticipate follow-up questions and address weak points before you even ask.

3. Coaching Questions

Ask AI to interview you rather than waiting for your instructions.

This is especially useful when you know you have a problem but haven't crystallized what you need.

Example Prompt "I'm struggling with something in my program but I'm not sure exactly what the real issue is. Act as a coach. Ask me up to 6 clarifying questions, one at a time, to help me articulate the problem before we discuss solutions."
Time Clarity Prompts (Dylan Davis)

Structured questions that help you articulate fuzzy goals. Instead of you trying to explain what you want, AI guides you through questions that surface what you actually need. Particularly powerful for complex challenges where the problem definition isn't clear.

The Four Roles AI Can Play

As you use the co-thinking techniques in Capture, Challenge, and Coach, AI can also be thought of as taking on different roles.

Strategist

Helps you see the big picture, identify priorities, and plan approaches

Critic

Challenges your thinking, finds weaknesses, and stress-tests ideas

Planner

Breaks down projects, creates timelines, and organizes tasks

Builder

Creates drafts, generates content, and produces deliverables

(Pro tip: You can even explicitly ask AI to take on these roles for emphasis.)

Every Phase of Your Work

Use AI throughout a project lifecycle—not just at specific moments. Planning, research, discussion, creation, feedback and testing.

Co-Working

Co-working is what happens when AI moves beyond conversation to actually doing work alongside you—autonomously executing tasks while you focus on higher-level guidance.

Co-Thinking
Interactive conversation. You and AI go back and forth. You're actively engaged throughout. The chat window has your full attention.
Co-Working
Autonomous execution. You give AI a goal and context. AI works independently. You review outputs when ready. Multiple tasks can run in parallel.

Currently available examples: Claude Cowork and Google Antigravity are early implementations of this co-working paradigm, allowing AI to work on tasks autonomously while you focus elsewhere.

Vibe Coding (and Vibe Everything)

"Vibe coding" is a term for describing what you want built rather than how to build it. You specify the outcome; AI handles the implementation. This extends beyond code: you can "vibe" your way to slide decks, research summaries, data analyses—any complex output where you know the goal but not every step.

The Inbox Future

The future of AI work isn't a chat window you stare at. It's an inbox you check.

You tell a team of AI agents what you want and provide ways for them to grade their work as successful. You don't need to tell them how. They work autonomously and report back when they're done or need input.

From Chat Box to Inbox

Imagine this workflow: You open your AI dashboard in the morning. Several tasks are in progress—a research summary on new ACGME requirements, a draft schedule optimization, talking points for the faculty meeting. You didn't watch any of these happen. You review the outputs, provide feedback or approve, and queue up new tasks.

For radiology: A few cases opened, each being worked on in parallel. You're not waiting and watching. You're managing a portfolio of AI-assisted work, checking in when outputs are ready.

Today

Interactive chat—your attention required

Emerging

Autonomous tasks—check back later

Future

Many agents working simultaneously — Agents check in via Inbox

What About Resident Training?

If AI handles low-level work, how do residents develop foundational skills? This question doesn't have easy answers, but it needs active engagement.

Expertise Through Repetition

Residency has historically relied on repetition of fundamental tasks to build expertise. Reading hundreds of studies builds pattern recognition. Writing countless reports develops communication skills. Working through problems step-by-step creates problem-solving intuition.

If AI can generate preliminary reads, draft reports, and walk through differential diagnoses, what happens to this developmental process? Are we removing valuable struggle, or freeing residents to focus on higher-order skills sooner? We've navigated similar transitions before—dictation software, PACS, search engines, electronic records—and each time we adapted training. This may be different in scale, but perhaps not in kind.

Platform Quick-Start Guides

Specific instructions for getting started with the major AI platforms and their collaboration features.

Claude (Anthropic)

claude.ai

Claude Projects (Second Brain)

Projects let you create persistent workspaces with uploaded documents that Claude references across conversations.

1

Create a Project

Click "Projects" in the sidebar → "Create Project" → Name it (e.g., "Program Director Resources")

2

Add Knowledge

Upload PDFs, documents, or paste text into the project's "Knowledge" section. Claude references these in all conversations.

3

Set Custom Instructions

Add instructions that apply to all conversations (e.g., "You're helping a radiology program director. Always consider ACGME requirements.")

Claude Projects Documentation →

Extended Thinking (Reasoning Mode)

Claude can show its reasoning process for complex problems.

1

Enable Extended Thinking

Click the lightbulb icon or toggle "Extended thinking" before sending a complex query.

2

Review the Thinking

Claude shows a "Thinking..." section where you can see its reasoning before the final response.

Claude for Chrome (Browser Agent)

Claude can control a browser to navigate websites, fill forms, and complete web-based tasks.

1

Install Extension

Search "Claude for Chrome" in Chrome Web Store and install the official Anthropic extension.

2

Direct Browser Tasks

Ask Claude to "go to [website] and find [information]" or "fill out this form." Review actions before confirming.

ChatGPT (OpenAI)

chatgpt.com

ChatGPT Projects

Create workspaces with persistent files and custom instructions.

1

Create a Project

In the sidebar, click "Projects" → "New Project" → Name and describe your project.

2

Upload Files

Add documents to the project. ChatGPT uses these as context for all conversations.

ChatGPT Projects Documentation →

Voice Conversations

Have natural spoken conversations using the mobile app or web interface.

1

Mobile App

Tap the headphone icon to start a voice conversation. Speak naturally; ChatGPT responds aloud.

2

Web Voice

Click the microphone icon to dictate. Click the speaker icon on responses to hear them read aloud.

Custom GPTs

Create specialized AI assistants with specific instructions and knowledge for recurring tasks.

1

Create a GPT

Click "Explore GPTs" → "Create" → Describe what you want your GPT to do.

2

Add Knowledge & Share

Upload documents in "Configure" tab. GPTs can be private, shared via link, or published publicly.

Custom GPTs Documentation →

Google AI Tools

Gemini & NotebookLM

Google Gems (Custom Personas)

Gems are custom versions of Gemini with specific instructions for different use cases.

1

Access Gems

In Gemini (gemini.google.com), click "Gem manager" in the sidebar.

2

Create & Use

Click "New Gem" → Give it a name and detailed instructions → Select it from sidebar to use.

Google Gems Documentation →

NotebookLM (Research Assistant)

Google's AI research tool that works only with sources you provide—no external training data mixed in.

Why NotebookLM is Different

Unlike general chatbots, NotebookLM only answers based on documents you upload. Excellent for research where you need AI to stay within specific sources without hallucinating from training data.

1

Create a Notebook

Go to notebooklm.google.com → "New Notebook" → Name your project.

2

Add Sources

Upload PDFs, paste text, add Google Docs, or link YouTube videos and websites.

3

Query & Generate

Ask questions of your sources. Generate study guides, FAQs, or audio overviews.

NotebookLM Documentation →

Browser Automation

AI-Controlled Web Navigation

What is Browser Automation?

Browser automation lets AI act as an autonomous agent that can navigate websites, "see" pages through screenshots, and click buttons to complete tasks on your behalf. These tools can research across multiple sites, fill out forms, and interact with web applications.

Available Tools

1

Claude for Chrome

Anthropic's browser extension lets Claude see your screen, navigate pages, and click elements to complete multi-step web tasks autonomously.

2

Gemini for Chrome

Google's extension enables Gemini to view and interact with web pages, taking screenshots to understand context and executing clicks and navigation.

3

Perplexity Browser Comet

Combines search with browser capabilities to research topics across the web, synthesizing information from multiple sources into coherent answers.

Start with a Problem. Try Co-Thinking.

It's a great time to start. You're not behind. It's exciting and easier than ever before.

01

Pick One Real Problem

Something you're actually working on this week. Not a test case—a genuine task where better thinking or faster execution would help.

02

Try One CAPTURE Technique

Voice dictation, second brain, or format transformation. Experience AI as capture tool before moving to collaboration.

03

Experiment with CHALLENGE

Ask AI to push back on something. Request non-obvious solutions. Have it play devil's advocate on a decision you're considering.

04

Let AI COACH You Once

For a complex problem, let AI ask you questions instead of the reverse. See what emerges when you're not driving.

"A portion of your attention, for the rest of your life, needs to be devoted to becoming a better collaborator with AI."
— Jeremy Utley

Ready for videos, newsletters, and more resources?

Keep Learning →