Clinically Intelligent is written by Juwon Akinyande in a personal capacity and is not affiliated with or representative of Barts Health NHS Trust or any other NHS organisation.
Content is intended for general educational purposes only and should not be treated as clinical, legal, or information governance advice. Local policies and governance requirements differ between organisations. Before applying any workflow or guidance in practice, clinicians should follow local policy and seek appropriate governance advice where needed.
Before I started this newsletter most of my experience with AI was in the rest of my life. Business. Organising things. Getting through the week. I assumed healthcare was different. I assumed healthcare was different. I assumed the restrictions were so tight and the integrations so limited that the only real use cases were note taking and a bit of research. I thought other industries were better suited to AI than healthcare even though healthcare is drowning in exactly the kind of repetitive work AI handles well. I thought the barriers were technical and the changes worth making were too big for the tools and the knowledge I had.
I also thought I needed to be a technical or security expert before I could widely use AI at work let alone tell anyone else how to.
The barrier is not the one you think it is.
I assumed the thing standing between me and using AI properly was technical knowledge. It was not. The thing standing in the way was not knowing where the line was. Once I had done the research on the governance side — what I could put in, what I could not, where the risk sat — the confidence followed almost immediately. That is the whole reason I built the Safety Check last week. Do that research once and you save yourself a huge amount of hesitation every time after.
You do not need to be a security expert. You need to know where the line is and then be willing to think.
More tools is not better.
I ran into walls. At the time I tested it, Granola required admin approval before I could even log in. OpenEvidence did not work in the UK. Small things but they added up to a pattern. Access in healthcare is its own problem.
But the bigger lesson came from a workflow I tried to build and could not. I wanted to take feedback from people and turn it into a first creative draft using a few tools together. The concept made sense in my head. In practice it fell apart. Not because the tools could not do it. Because when I imagined a clinician juggling three tools at once to save time I could see it doing the opposite. What is the point of using AI to save time if it raises your stress while doing it.
Just because AI can be used somewhere does not mean it should be. A few tools used well will always beat fifty tools used badly.
AI does not always save time.
The clearest example was that same workflow. At the time I tested it, I wanted to see whether Napkin could handle unstructured human feedback as well as it handled a clean document like my Issue 04 workflow. I gave it composite feedback shaped into a design brief. Napkin gave me options. One of them was a basketball metaphor — nice but not what I needed. None of them matched what I was trying to communicate. I tried again with the feedback laid out differently. Again. Again. Most of my time went into restructuring the feedback and hunting for an image that never came.
In the end I had spent more time going back and forth between ChatGPT and Napkin than it would have taken to make the design myself on Canva.
The productivity story says AI saves time. Sometimes it does. Sometimes the honest answer is that the human way was faster. Knowing which is which is the skill nobody talks about.
NotebookLM.
I have not used it to its full capacity yet and it has already changed how I work. The part that still surprises me is the audio overview. You feed it information from several sources and it produces an audio summary that sounds like two people having a conversation. It turns dense heavy material into something you can listen to on the go without your eyes glazing over.
It means I can learn while moving. A stack of documents or videos I would have struggled to get through becomes something I can absorb on a walk. That one feature has changed how much I can take in and how I take it in. It is the clearest example I have of a tool doing something genuinely better than I expected.
The thread running through all of it.
The failed workflow. The access issues. The Napkin afternoon. The NotebookLM surprise. Every outcome came down to whether I had thought clearly about what I was actually trying to do before I opened anything.
AI can help you think faster. It cannot think for you. The clinician who actually thinks about the problem will always get more out of these tools than the one who just learns the buttons.
When I started I thought this newsletter would mostly be about tools. Seven issues in I think it is about something else. The interesting question is no longer what AI can technically do. It is whether healthcare can use it safely, realistically, and without creating more friction than it removes.
Whether it works when it actually matters. Not in theory. On a Tuesday. When you have a full caseload, back to back meetings and no time for errors.
That is where this newsletter is heading.
Next week back to something practical. If someone you know would find this useful pass it on.
Clinically Intelligent drops every Wednesday. If you are not yet subscribed you can join free at clinicallyintelligent.com.
The AI tools discussed in Clinically Intelligent are consumer products. They have not been independently assessed by the author against DCB0129 or DCB0160 clinical risk management standards, and they may not be approved for clinical use by your employer.
Before using any tool or workflow in connection with clinical practice, clinicians should ensure its use aligns with local policy, governance requirements, and professional responsibilities.
Patient identifiable information should not be entered into consumer AI tools unless explicitly permitted within an approved organisational workflow.
Examples and scenarios discussed throughout the publication are fictional and composite unless otherwise stated.