Think Like a Clinician, Build Like a Developer: How to Build Your Own AI-Powered Tools

Healthcare professionals encounter workflow challenges every day, stemming from operational inefficiencies, duplicative processes, and gaps that no single product quite solves. Traditionally, addressing these problems required IT teams, funding, and time. Now that’s changing.

With the rise of AI-assisted development, often called “vibe coding”, clinicians can now build simple, useful tools themselves by describing what they need in plain language.

This shift is powerful: it places innovation directly in the hands of clinicians, enabling faster, more relevant solutions tailored to real-world clinical workflows.

Getting Started:
One of the simplest ways to approach tool development is to follow an iterative cycle.

1. Define the Problem (Not the Technology)

• Start with a workflow challenge, not a tool idea.
• What is slow, repetitive, or frustrating?
• What outcome do you want instead?
• Start small, with workflows that do not contain  patient-specific information

2. Describe the Solution in Plain Language
AI coding tools allow you to build using natural language. To ensure success, translate your idea into a clear structure. This should include:

• Core function
• Key features
• Inputs and outputs
  • Break down the problem in steps to identify where it becomes complex – think: How would I do it?
• Constraints (e.g., make sure the tool works only on your device and does not share or send any info externally, for privacy and security)

3. Start to Build: Using AI-Enabled Tools
Platforms such as Cursor or Windsurf* allow you to generate code automatically, rapidly prototype apps, and iterate quickly without deep technical knowledge. Many platforms offer free trials to get started!

• Ask the LLM to provide the solution in a step-wise manner and with a way to test that each step works.
• The bigger the ask, the greater the risk of inaccuracy. By breaking it into steps, the LLM task scope is reduced, and testing it after each step makes it easier to identify what’s not working and what to debug.

These tools act as development assistants, helping users build and refine applications much faster than traditional methods. The clinician should always remain the independent decision-maker and maintain oversight.

Example prompt: “Build an offline, easy-to-run app that allows me to convert Standard SI (Canadian) for the most common 15 medical lab tests.”

As the tool works, you can monitor and verify its processes and planning and stop it if you feel it’s not on track (how you do this will vary by platform).

4. Review and Validate the First Draft
Treat the initial output as a prototype, not a finished product.

• Check for accuracy and usability
• Ensure it aligns with your intended workflow
• Identify gaps or errors

Remember: AI is assisting, not deciding.

5. Test with Safe Data
Always test using:

• Synthetic (fake) data
• De-identified data
• Non-sensitive workflows

This helps you validate your tool without exposing personal or patient information.

6. Refine Through Feedback
Improvement happens through iteration.

• Provide plain-language feedback to the AI
  • Example: “I noted that your output wasn’t correct for this, can you check it?”
• Break problems into smaller parts if needed
• Adjust and rebuild incrementally
• Nothing with personal/patient information (PI/PHI) should go back into the LLM/coding environment

Privacy and Safety Guardrails

AI introduces new risks if used improperly. While building your own tools is increasingly accessible, privacy and security must always come first.

Key Principles to Follow

• Follow guidance from your regulatory college
• For physicians, additional guidance is available at the Canadian Medical Protective Association (CMPA)
  • Considerations for adopting and implementing AI in Healthcare
  • CPSO – Using AI in Clinical Practice
• Where applicable, also follow your organization’s IT, privacy, and security processes
  • Check before downloading trial or other new tools on a work device
  • Avoid using your work email or work login with unapproved tools (even on your personal device)
  • Get organizational sign-off before using an AI tool in routine workflow
• Do not input Personal Information or Personal Health Information into general-purpose AI/coding tools
• Use de-identified or synthetic data during development
  • Reduces risk of unintended exposure of PI/PHI
• Minimize data input
  • Use the minimum necessary information for any task
• Set up your tool’s privacy/security settings so that you can process data locally, not using a vendor’s cloud
  • Ensures information does not leave your control
  • Be cautious with free or trial tools, which may offer weaker privacy controls
• Maintain strict clinician oversight
  • AI outputs must always be reviewed before use, with strict clinician oversight

Amplify Care recently hosted a webinar where two family physicians, Dr. Byron Song and Dr. Kevin Brophy, shared how they are using coding tools to address administrative and workflow challenges in primary care

Anyone interested in exploring how these tools are being applied in real-world primary care settings can access the webinar recording here.

* Please note that the vendors mentioned are provided solely as illustrative examples of service providers within this industry. Reference to these companies does not constitute an endorsement, recommendation, or vetting of their specific products or services.