Conversations with AI begins with a question: how can working with these systems help us develop our own capacities?
The project draws on research into how people assess their understanding, how AI responds to personal experience, and what people need from technology entering the work of care. These studies inform the approach; the encounters themselves are where we begin to discover what it can offer. Lee et al., 2025, Basar et al., 2025, Cecil et al., 2026.
Suppose you’re writing a short introduction to your practice. You have some rough notes, but the words aren’t coming together. You ask AI to help.
Seconds later, there’s a paragraph. It’s clear, confident and considerably more polished than what you started with. You could probably use it.
Then you read it again. Would you actually say that? Does it describe how you work, or how someone in your profession is expected to sound? Perhaps it names something you’ve struggled to express. Perhaps you like the confidence more than you recognise the person.
You might keep most of it, change a sentence or decide your rough notes were closer. What interests me is how you arrive at that decision.
Bringing another kind of attention
My work as a therapist involves listening beyond the first explanation: noticing hesitation, attending to feeling and bodily experience, and asking questions that can make an otherwise familiar situation available to reconsider.
I want to bring that depth of attention to our encounters with AI. A response can be factually sound and still miss what matters to someone. A useful suggestion can uncover an intention they hadn’t found words for. Understanding these moments takes more than knowing how to write a prompt.
Some of my own exchanges with AI have been difficult to convey afterwards. They have left me persuaded that there is something here worth pursuing with considerable care and ambition. I want to understand the conditions under which this kind of exchange becomes genuinely fruitful, and help more people find their way into it.
That is also a design problem. What should software make easy? When should it invite another look? How can it offer substantial assistance while leaving someone actively involved in the thinking?
Working on something together
Conversations with AI is taking shape through an initial one-to-one pilot with wellbeing practitioners. I’m there from the first encounter. We work on a practical situation, with room to notice what happens along the way.
Imagine drafting a reply about a cancellation fee. AI produces something courteous and firm. The wording seems reasonable, but sending it still feels difficult.
Perhaps the difficulty is in the wording. Perhaps you haven’t decided what boundary you want to hold. Perhaps the draft gives you a starting point for making that decision. We can explore those possibilities while continuing to work on the reply.
I’m developing the sessions through rehearsals of recognisable situations, followed by real encounters. What happens in those conversations will guide the software: its structure, the questions it offers and the choices it leaves with the person using it.
Where this could lead
The eventual software could help someone describe their work in a voice they recognise, prepare a workshop, navigate an awkward message or turn an overwhelming task list into a manageable next step. It could support closer examination of an evidence claim, a seemingly insightful reflection, or a convenient summary that has lost the dissenting voices. For practitioners, it could also provide a place to consider how to respond when someone brings AI-generated advice into their work together.
The wider ambition is to make thoughtful use of AI more accessible. People should not need technical expertise to obtain useful assistance, question its assumptions or recognise when their own knowledge should lead. If this work succeeds, the benefit could extend beyond individual productivity: more time available for care and teaching, better-informed decisions, and more people able to participate in shaping how these technologies enter everyday life.
Research informing the project
- Lee et al. (2025). Learning Behaviors Mediate the Effect of AI-powered Support for Metacognitive Calibration on Learning Outcomes. A randomized educational study connecting structured support for self-assessment with learning gains.
- Basar et al. (2025). How Well Can Large Language Models Reflect?. An evaluation of AI-generated reflections in motivational-interviewing dialogues, examining both their promise and limitations.
- Cecil et al. (2026). Navigating the complexity of AI adoption in psychotherapy by identifying key facilitators and barriers. Research with therapists and patients highlighting the importance of human contact, practical usefulness and the context of adoption.