I'm Joseph Fisher. I've spent over a decade working in coaching, consultancy, and organisational systems where working with complexity is the whole point.
Social impact, public services, organisational change. The kind of work where the most important things are often not obvious and the diversity of perspectives and the conversations people have are what matters the most.
When I first started using large language models, I wanted something that could think with me. But most of the time the AI just wanted to turn everything into a neat and coherent summary and resolve the most important and interesting parts into excitable bullet points.
There was no nuance, no honouring of paradox and emergence, no awareness of context, no acknowledgement of what wasn't known. So I started experimenting. Finding ways to work with the models so they got better at holding complexity rather than flattening it.
I started with prompting and talking differently to the models, and then built knowledge structures that captured how the ideas and concepts I use relate to each other. I worked out how to design reasoning tools that followed the thinking processes I actually use - not generic "think step by step" instructions, but constraint architectures drawn from real methodology that genuinely support deeper inquiry and reflection.
Some colleagues saw the results I was getting and asked if I would make them their own tools. That's how Bespoked AI started, and it's still how most of the work begins - someone sees what this can do and wants it applied to their domain.
