I use AI in developing ideas and producing much of the work I publish. It is part of how I explore questions, make connections, organise material, draft and revise. I want readers to understand what that involves, including what AI contributes and what I remain responsible for.

My work draws on around thirty years as a clinician, researcher, academic, consultant and board member. Over that time I have accumulated substantial amounts of research, observations, frameworks and questions about how health systems work. AI helps me bring that material together and develop it into something I can share and others can examine.

The phrase I use for this way of working is ‘thinking in partnership’.

Developing ideas through conversation

A September 2026 essay by philosopher Joe Cruz in The New York Times resonated with me. He questioned the assumption that writing is the primary route to developing thought, and made a case for conversation and the ability to explain and defend ideas. I recognised something of my own way of working in that argument.

Writing is, and has always been, important to me. I keep a daily journal, and I use that to seed, explore and curate my thoughts and ideas.  I also use conversation as an important thinking tool. Explaining an idea, hearing a response and working out why it does not quite fit can make the idea clearer. Discussing an idea often changes what I think altogether. This is an approach I use in all my work, through facilitated workshops and co-design methodologies.

AI gives me a useful way to have that exchange without a human audience. I might begin with something I have written, a question from a meeting, an article that connects with earlier work, or a few minutes of actually talking through a problem. I bring the background, explain what bothers me, and explore what comes back. Because I use structured folders within AI, there is normally deep context from my previous work to build on.

The useful work happens through the exchange. I add context, question an interpretation, ask for an alternative explanation or bring in evidence that complicates the answer. The question itself often changes as the conversation evolves.

AI sometimes contributes a connection or a way of framing a problem that I had not arrived at myself. That contribution can be substantial. I test whether it holds against the evidence and my experience, and decide whether to pursue it. The process helps develop the argument as well as express it. As an academic, I particularly value the interdisciplinary input and context that AI can provide. This often moves me outside my familiar disciplinary traditions and lets me see how my work interacts with other disciplines, theorists, academics and ideas.

Turning that thinking into writing

The tools I use on a regular basis are ChatGPT and Claude. I use them to explore ideas, work across existing material, suggest structures, draft passages and revise text. Sometimes I ask AI to interview me about a specific topic, then work with it to turn my answers into a draft.

The resulting prose always needs attention. AI can smooth away an important distinction, remove an awkward but necessary qualification, or make an argument sound more certain than the evidence allows. It can also produce perfectly competent sentences that are not accurate or do not sound like me. I push back, restore material, rewrite and reconsider.

A fast draft can still need considerable time. My writing changes through feedback from readers, changes in my own understanding and repeated decisions about what belonged in it. Producing text and deciding that it is ready to publish are different parts of the work.

Evidence and responsibility

AI can be wrong, including when it sounds persuasive. It can also be too agreeable. A response that develops my argument smoothly may still leave its weakest assumption unexamined. I need to question both the answer and the way I have framed the problem. My academic training has taught me how to both frame questions, and interrogate the quality and rigour of the findings. I bring both disciplines into my writing with AI.

I treat factual claims, quotations and references produced by AI as things to check against original sources. Asking a model to confirm its own answer is not independent verification. My professional experience helps me recognise problems, but it does not make me immune to errors or confirmation bias.

Human discussion and criticism remain part of the process. Increasingly, I use active networks and communities of practice to develop, test and iterate ideas in partnership with those communities. In every case, I have found that the addition of more perspectives and human input has changed, and generally improved the outputs. However, AI can provide useful scaffolding as a basis to start the discussions. An AI conversation cannot establish on its own that an argument will survive scrutiny beyond that conversation.

Confidentiality also matters. Having access to information does not automatically give me permission to put it into an AI system. The material, the setting and any obligations to other people have to guide those decisions. Where I write about people who could be identified, I ask them first.

I attribute ideas to the people they came from, even when AI has helped me phrase them. AI can reproduce other people’s thinking without saying where it came from, so I check. The frameworks I write about are my own work, developed over many years. AI has helped me test and refine them. It did not originate them.

Some of my work is subject to specific rules about AI use, including peer review, examinations and expert evidence. Commissioned work is governed by my agreement with the client. Where these requirements apply, they take precedence over the general approach described here.

Some images on this site are AI-generated and labelled as such.

When I publish something under my name, I take responsibility for its argument, its use of evidence and the final text. AI involvement does not transfer that responsibility. If I have made an error, I welcome being told and will correct it. You can contact me here.

Why I work this way

AI makes it easier for me to connect work that previously sat in separate projects and to bring more of it into public discussion. It also makes it very easy to produce more than is useful. I still have to decide which questions deserve attention and whether a piece of writing has something worthwhile to contribute.

I am making this process explicit because readers should know how the work they are engaging with has been developed. This article was prepared with AI assistance too. I invite you to assess what I publish through its reasoning, the evidence it draws on and whether it helps you understand or question something that matters.

Last updated: October 2026.