A human cognition layer inside an AI-assisted design workflow.
HMHT applies human-centred design discipline to the earliest stages of the SDLC, using AI to simulate how a specific, well-defined user would reason, where direct research access is constrained. What's here is the position and the proof. The exact mechanics stay close for now.
Ask an AI system to go straight from a brief to a screen, and it will. What it won't do on its own is reason through how a specific, real person would actually approach the problem. Skip that step and AI-generated design looks plausible without being grounded in anyone's actual reasoning.
HMHT exists to put that reasoning step back in, deliberately, before generation happens. Every simulated output it produces is treated as a hypothesis to validate against real users, never a substitute for them.
HMHT isn't one-size-fits-all. It's been run end-to-end on some engagements, and scoped to specific decisions inside a larger process on others.
This page shows what HMHT is for, and where it's been proven. It doesn't show the exact structure underneath, the specific stages, the prompting patterns, or how simulated reasoning gets validated before it's trusted. That's still being formalised into something meant to be taught more broadly.
Every case study this method touches shows the outcome, and increasingly, like Goodwill, shows the method in action too. This page draws a different line: it shows HMHT works and has range, without handing over the exact mechanics. A working method is worth more applied in a real conversation than described in full on a public page.
Want to see the full methodology, phases, prompting architecture, and all? I'm happy to walk you through it directly.
Request a walkthrough